Professional Documents
Culture Documents
‡
Princeton University, Department of Mathematics,
Fine Hall, Washington Road, Princeton, NJ 08544, United States of America
§
University of Cambridge, Department of Pure Mathematics and Mathematical Statistics,
Wilberforce Road, Cambridge CB3 0WA, United Kingdom
∗
Westfälische Wilhelms-Universität Münster, Mathematisches Institut,
Einsteinstrasse 62 48149 Münster, Bundesrepublik Deutschland
28 December 2022
Abstract
We prove global existence and decay for small-data solutions to a class of quasilinear wave equations
on a wide variety of asymptotically flat spacetime backgrounds, allowing in particular for the presence
of horizons, ergoregions and trapped null geodesics, and including as a special case the Schwarzschild
and very slowly rotating |a| M Kerr family of black holes in general relativity. There are two
distinguishing aspects of our approach. The first aspect is its dyadically localised nature: The nontrivial
part of the analysis is reduced entirely to time-translation invariant rp -weighted estimates, in the spirit
of [DR09], to be applied on dyadic time-slabs which for large r are outgoing. Global existence and
decay then both immediately follow by elementary iteration on consecutive such time-slabs, without
further global bootstrap. The second, and more fundamental, aspect is our direct use of a “black
box” linear inhomogeneous energy estimate on exactly stationary metrics, together with a novel but
elementary physical space top order identity that need not capture the structure of trapping and is
robust to perturbation. In the specific example of Kerr black holes, the required linear inhomogeneous
estimate can then be quoted directly from the literature [DRSR16], while the additional top order physical
space identity can be shown easily in many cases (we include in the Appendix a proof for the Kerr case
|a| M , which can in fact be understood in this context simply as a perturbation of Schwarzschild).
In particular, the approach circumvents the need either for producing a purely physical space identity
capturing trapping or for a careful analysis of the commutation properties of frequency projections with
the wave operator of time-dependent metrics.
Contents
1 Introduction 3
1.1 Assumptions on (M, g0 ): “black box” estimates and physical space identities . . . . . . . . . 4
1.2 Assumptions on the nonlinearities g(ψ, x) and N (∂ψ, ψ, x) . . . . . . . . . . . . . . . . . . . . 9
1.3 The dyadically localised approach to global existence: sketch of the proof of the main theorem 10
1.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.5 Outline of the paper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.6 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1
2 Geometric assumptions on the background spacetime 17
2.1 Underlying differential structure and the positive function r . . . . . . . . . . . . . . . . . . . 17
2.2 The Lorentzian metric g0 , time orientation and the causality of the boundary . . . . . . . . . 18
2.3 The stationary Killing field T . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4 The Killing horizon H+ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.5 Foliations, subregions and volume forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.6 Other vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.7 Examples: Minkowski, Schwarzschild and Kerr . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.8 Table of r-parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2
1 Introduction
We consider here quasilinear equations of the form
on a 4-dimensional manifold M, where g(ψ, x) and N (∂ψ, ψ, x) are appropriate nonlinear terms and where
g(0, x) = g0 (x) defines a stationary asymptotically flat Lorentzian metric on M foliated by a suitable family
of hypersurfaces Σ(τ ), for τ ≥ 0, which for large r are outgoing null. In this paper, we prove the following:
Theorem. For equations (1.1), under appropriate assumptions on the background spacetime (M, g0 ) and
on the nonlinearities g(ψ, x) and N (∂ψ, ψ, x) to be described below, we have
• Global existence of small data solutions. Solutions ψ arising from smooth initial data on Σ0 ,
sufficiently small as measured in a suitable weighted Sobolev energy, exist globally on M.
• Orbital stability. Under the above assumptions, the above weighted energy flux through Σ(τ ) is
uniformly bounded by a constant times its initial value.
• Asymptotic stability. Under the above assumptions, a suitable lower order unweighted energy flux
through Σ(τ ) decays inverse polynomially in τ (implying also pointwise inverse polynomial decay for ψ).
One possible motivation for studying (1.1) is as an illustrative model problem for issues relating to the
nonlinear stability of the spacetimes (M, g0 ), when these themselves are solutions to the celebrated Einstein
equations of general relativity. An important example of such spacetimes (M, g0 ) allowed by our theorem is
provided by the Schwarzschild family of metrics for M > 0, and the more general Kerr family (parametrised
by a and M ) in the very slowly rotating regime |a| M . For a discussion of these spacetimes and the
stability problem in general relativity, see [DHRT21].
Among other features, the Schwarzschild and Kerr spacetimes exhibit trapped null geodesics, along which
energy can concentrate over large time-scales. Moreover, these are asymptotically flat spacetimes, meaning
that linear waves ψ decay like ∼ r−1 along outgoing null cones. This then constrains the decay of ψ in the
near region to also be at best only inverse polynomial. (In the above black hole examples, this slow decay
in the near region is already a linear effect due to scattering off of far away curvature associated to the
nontrivial spacetime mass at infinity; in Minkowski space, this slow decay in the near region arises due to
purely nonlinear scattering effects, even for compactly supported initial data.) It is the combination of these
two specific analytical difficulties—trapped null geodesics and the relatively slow decay in the near region
necessitated by asymptotic flatness—of the black hole stability problem (and related problems) which we
wish to capture with our assumptions in the present paper. It is for this reason that we include both the
nonlinear term implicit in the expression g(ψ,x) ψ on the left hand side of (1.1) (the “quasilinearity”), as
this is the most dangerous term in the vicinity of trapping, as well as the nonlinear term N (∂ψ, ψ, x) on the
right hand side (the “semilinearity”), as this term models the true null structure at infinity of the Einstein
equations, when the latter are written in appropriate geometric gauges. To avoid inessential complications,
we will in fact assume that g(ψ, x) = g0 for large r and that N (∂ψ, ψ, x) satisfies a generalised version of the
null condition [Kla86].
In the case where (M, g0 ) is Minkowski space, a version of the above theorem follows from classical
work of [Kla86, Chr86], and there have been many further amplifications over the years, particularly in
the context of the obstacle problem, see e.g. [MS05]. Concerning specifically the Schwarzschild and very
slowly rotating |a| M Kerr setting, versions of the above theorem have been shown previously in various
special cases; see already [Luk13] in the semilinear and [LT16, LT20] in the quasilinear case, as well as the
recent [LT22] (the latter three works concerning slightly different classes of equations, but satisfying the
weak null condition [LR03]). For axisymmetric solutions to certain semilinear equations on Kerr in the full
subextremal range |a| < M , see [Sto17]. See also [Pas19] for a related physically motivated quasilinear
problem and [HV16, Mav21] for the study of (1.1) on the non-asymptotically flat Schwarzschild–de Sitter
and Kerr–de Sitter black hole backgrounds, where the cosmological constant Λ is positive and decay of ψ is
in fact exponential. For work specifically connected to the related problem of stability of these spacetimes
themselves under the Einstein equations, see already Section 1.4.4.
In comparison to previous related work, there are two distinguishing features of the present approach.
3
The first distinguishing feature is our purely dyadic framework: Rather than relying on a global bootstrap
based on time-weighted norms, the argument is entirely reduced to rp -weighted but time-translation invariant
estimates to be applied in spacetime slabs of time length L which are outgoing null for large r. Global
existence (and decay) is then inferred by proceeding iteratively in time in consecutive slabs of length L = 2i ,
in the spirit of [DR09]. No bootstrap is necessary for the iteration itself, and to estimate the i’th slab,
no information is necessary to remember about the past other than information on the data of the slab
itself. In this sense, the argument is truly dyadically localised. This yields a more streamlined proof
even restricted to the semilinear case. (In fact, for both the semilinear and quasilinear cases, if (M, g0 ) is
a suitably small perturbation of Minkowski space, the time-translation invariant estimate can be applied
directly without iteration, and the approach proves global existence without explicit reference to time decay,
cf. the recent [FM22]. In our formalism we shall always assume g0 to be stationary, though we note that
results can also be obtained on non-stationary perturbations of Minkowski space [Yan13].)
The second, and more fundamental, feature is our direct use of a “black box” linear inhomogeneous energy
estimate (which in applications captures both complicated trapping phenomena as well as low frequency
obstructions) on the exactly stationary background, together with a physical-space top-order identity which
may be applied directly to the quasilinear equation (1.1) and is in fact completely insensitive to trapping
(and, when it holds, robust to perturbation of the metric g0 ). In the example of Kerr, in fact for the full
subextremal case |a| < M , our black box assumption follows directly from [DRSR16], while we show explicitly
how to retrieve the additional top-order physical space estimate in the case of |a| M (which in this context,
can in fact be thought of simply as a perturbation of Schwarzschild) in Appendix A. The correct formulation
of the companion estimate to be used in connection with [DRSR16] for the full subextremal case |a| < M will
appear elsewhere. (We emphasise that it is the phenomenon of superradiance which is the primary difficulty
in producing this identity, not the complicated structure of trapping per se; for instance, the necessary top
order physical space identity can be shown for general stationary spacetimes without horizons or ergospheres,
irrespective of the structure of trapping.) The approach thus does not depend on whether or not there exists
a purely physical-space based identity capturing trapping (something very fragile!) nor does it require a
careful analysis of the commutation properties of frequency projections with the wave operator g(ψ,x) of
the time-dependent metric g(ψ, x) corresponding to the actual solution ψ (something quite technical, which
was successfully done, however, for instance in [LT20]).
We emphasise that the role of the companion physical-space estimate is purely in order to deal with
quasilinear terms. When our method is restricted to the semilinear case, i.e. when g(ψ, x) = g0 identically
in (1.1), then the black box linear inhomogeneous energy estimate can be applied on its own. In particular,
in view of [DRSR16], our main theorem immediately applies to semilinear equations on Kerr in the full
subextremal range |a| < M .
We note that the direct appeal to a linear “black box” statement is similar in spirit to [MS05] for
instance, where results on quasilinear equations outside obstacles in Minkowski space were studied under an
exponential decay assumption concerning the linear problem.
The remainder of this introduction is structured as follows: In Section 1.1 below, we will discuss in
more detail the assumptions we make on the background (M, g0 ), introducing both the black box linear
inhomogeneous estimate and the additional physical space identity, followed by the assumptions on the
nonlinearity in Section 1.2. We will then sketch our proof of the above theorem in Section 1.3, introducing
our purely dyadic approach. Finally, we shall end in Section 1.4 with a general discussion, giving various
extensions of the method, describing in more detail the relation with the nonlinear stability problem for
black holes, and comparing in particular with the case of backgrounds modelled on Schwarzschild–de Sitter
and Kerr–de Sitter spacetimes and with the recent work [Mav21].
1.1 Assumptions on (M, g0 ): “black box” estimates and physical space identities
We will make assumptions directly on the geometry of (M, g0 ) and on properties of the linear inhomogeneous
wave equation
g0 ψ = F (1.2)
on the fixed background g0 .
4
Σ(τ1 )
S R(τ0 , τ
Cv
Σ(τ0 ) 1)
H+
r=R
Σ(0)
Basic degenerate integrated local energy estimate. Our fundamental “black box” assumption at the
level of solutions of the inhomogeneous linear equation (1.2) will be the validity of an integrated local energy
estimate
Z τ1 (−1−δ)
Z τ1 Z Z
χ 0 0 µ
F(v), E(τ ) + c E (τ )dτ 0 + c E
−1
0 0
(τ )dτ 0
≤ λE(τ 0 ) + C |(V0 µ∂ ψ + w 0 ψ) F | + C F 2,
τ0 τ0 R(τ0 ,τ1 ) R(τ0 ,τ1 )
(1.3)
where E(τ ) denotes an energy flux through the hypersurface Σ(τ ), F(v) denotes an energy flux through
C v ∩R(τ0 , τ1 ), λ ≥ 1, C > 0, c > 0 are constants, and V0 , w are a fixed vector field and function, respectively,
and τ0 ≤ τ ≤ τ1 are arbitrary. In Minkowski space, such estimates go back to Morawetz [Mor68]. The energy
flux E(τ ) will control all first order derivatives of ψ on the spacelike part of Σ(τ ) while it will only control
all tangential derivatives on the null parts of Σ(τ ); similarly, F(v) will control all tangential derivatives
on C v . The fluxes will also control a zeroth order term with weight r−2 . The two additional terms on the
(−1−δ)
left hand side are as follows: The quantity χ E 0 (τ 0 ) denotes an integral over Σ(τ 0 ) controlling all first order
derivatives of ψ, whose density is multiplied by χr−1−δ , where 0 ≤ χ ≤ 1 denotes a function which is allowed
to degenerate but must satisfy χ = 1 in a region to be discussed below, and where δ > 0 can be taken to be
5
an arbitrarily small constant which will be fixed throughout. The quantity −E1 0 (τ 0 ) denotes simply the zeroth
order quantity Z
0
E
−1
(τ ) := r−3−δ ψ 2 .
Σ(τ )
The subscript −1 indicates that it is of order 1 less in differentiability than the other energies considered.
We recall that in our setup, what we define to be r satisfies r ≥ r0 > 0, and thus r-weights are only relevant
as r → ∞.
In the case where M has no spacelike boundary, i.e. S = ∅, and T is globally timelike, our only assumption
on χ will be that χ = 1 for large r. In the case where S 6= ∅, then we will require that χ = 1 for r ≤ r2 for
some r2 > rKilling . See already Section 3.2 for the detailed assumptions.
Let us note that estimate (1.3), with χ satisfying the above properties has indeed been shown in [DRSR16]
in the Kerr case for the full subextremal range of parameters |a| < M . One essential feature in deriving (1.3)
in Kerr is the fact that the trapped null geodesics of g0 , corresponding to photons in bound orbits around the
black hole, are unstable in a suitable sense. We note that one can show in general that estimate (1.3) in the
above form cannot hold in the presence of stable trapping. (Thus, (1.3) in particular constrains properties
of geodesic flow which can be expressed purely geometrically in terms of g0 .) Similarly, the possibility of an
estimate (1.3) with χ = 1 in a neighbourhood of H+ is related to the non-degenerate property of the horizon
of subextremal Kerr and its celebrated local red-shift effect. See [DR13].
Physical space energy identities. One way to try to prove (1.3) is through a physical space energy
identity arising from integrating the divergence identity
for well chosen currents J V,w,q,$ [ψ, g0 ] (associated to a vector field V , a scalar function w, a one-form q and
a two-form $) with (degenerate) coercivity properties for the arising bulk and boundary terms. In this case,
if one defines Z Z
E(τ ) = JµV,w,q,$ [ψ]nµΣ(τ ) , F(v) = JµV,w,q,$ [ψ]nµC ,
v
Σ(τ ) C v ∩R(τ0 ,τ1 )
then one has E(τ ) ∼ E(τ ), F(v) ∼ F(v), and one can reexpress (1.3) with E(τ ), E(τ0 ) replacing E(τ ), E(τ0 ),
respectively, but with λ in (1.3) now equal to 1. This is precisely the situation in the Schwarzschild case,
where appropriate coercive purely physical space identities of the form (1.4) can be deduced from [DR07a,
MMTT10].
If (1.3) indeed holds as a consequence of the coercivity properties of a divergence identity (1.4), then
this situation appears very advantageous for nonlinear applications. The advantage of such coercive iden-
tities (1.4) is that they are robust and easily amenable to direct application to the quasilinear (1.1). This
is because the latter can be viewed as an inhomogeneous wave equation with respect to the wave operator
g(ψ,x) corresponding to the solution itself, and both the existence of energy identities like (1.4) and their
coercivity properties are stable to passing from g0 to g(ψ, x), i.e. still hold for J V,w,q,$ [g(ψ, x), ψ].
If one only has estimate (1.3) as a “black box”, i.e. not proven via a physical-space identity (1.4), one
may still of course apply the estimate to (1.1) by rewriting (1.1) as an inhomogeneous equation with respect
to g0 :
g0 ψ = (g0 − g(ψ,x) )ψ + N (∂ψ, ψ, x). (1.5)
Applied in this manner, however, the resulting estimate loses derivatives, in view of the quasilinear second
order terms on the right hand side of (1.5). (Note in contrast that in the purely semilinear case, where
g(ψ, x) = g0 , there is no such loss in estimating (1.5), and we may base our argument entirely on (1.3). See
already Section 1.4.1.) At first glance, this loss would appear to present a fundamental difficulty.
An auxiliary physical space estimate. Here we come to the key observation of the present approach:
To avoid the loss of derivatives described above, one does not need that (1.3) be proven via a physical space
identity (1.4); one only needs a much weaker (from the coercivity point of view) estimate arising from (1.4)
which need not imply (1.3) but can be used in conjunction with (1.3).
6
Specifically, our main assumption is that, in addition to (1.3), one has a physical space identity (1.4)
with (a) coercive boundary terms and with (b) bulk terms which are only assumed to be nonnegative at
highest order, i.e. up to lower order “error” terms. Importantly, however, these allowed lower order error
terms must moreover be supported entirely in the region where the degeneration function χ of (1.3) satisfies
χ = 1, i.e. where the estimate of (1.3) is in fact non-degenerate. The identity (1.4) applied to (1.2), on its
own, gives rise thus to an estimate of the form
Z τ1 Z τ1 (−1−δ)
Z τ1 ξ Z
0 0
F(v), E(τ 0 ) + c ρ 0 0
E (τ )dτ 0 + c ρ 0 0
E
−1
(τ )dτ 0
≤ E(τ0 ) + A E
−1
(τ )dτ 0
+ |(V µ ∂µ ψ + wψ) F | ,
τ0 τ0 τ0 R(τ0 ,τ 0 )
(1.6)
(−1−δ) (−1−δ)
where ρ E 0 is defined similarly to χ E 0 but with degeneration function ρ which in general vanishes on a bigger
set than χ. The zeroth order term ρ−E1 0 on the left hand side also degenerates, unlike the analogous term
in (1.3). The presence of the term multiplying the new constant A on the right hand side is necessary in
ξ
view of the fact that nonnegativity is only assumed for the highest order terms. Here, −E1 0 is a zeroth order
energy whose density is supported only in the support of a function ξ(r)
ξ
Z
0
E
−1
(τ ) := ξ(r)ψ 2 .
Σ(τ )
Importantly, in view of our comments above, we will require that ξ vanishes identically where χ of (1.3)
degenerates, i.e. ξ = 0 where χ 6= 1. We will also require ξ to vanish for large r, and, in the case where
S= 6 ∅, in the region r ≤ r2 .
In addition, we will make on ρ the same asymptotic non-vanishing assumption that we we made previously
on χ, namely that ρ = 1 for large r, and, in the presence of a nonempty spacelike boundary S = 6 ∅, that
ρ = 1 in r ≤ r2 . See already Section 3.4.3 for the detailed assumptions and Figure 2 for a depiction of the
supports of χ, ρ and ξ.
We note that in the case where S = ∅ and T is globally timelike, one can in fact easily construct a current
satisfying (a) and (b) and leading to (1.6), irrespective of the structure of trapping and the validity of an
estimate of the form (1.3). We will show in Appendix A that for the very slowly rotating Kerr case |a| M ,
although there is no globally timelike T , there is a similar elementary construction, based essentially only
on the fact that superradiance is effectively governed by a small parameter and the ergoregion is confined to
a part of spacetime which can be understood using only the red-shift effect (cf. the original treatment of the
boundedness problem on Kerr [DR11]). In general, however, the main difficulty in proving (1.6) is ensuring
the boundary coercivity (a).
Though on its own, estimate (1.6) is clearly weaker than (1.3), its robustness to direct application
to (1.1) allows one to avoid the above loss of derivatives, when (1.6) is used in conjunction with (1.3) applied
to (1.5). Before turning to describe how this is done in the context of the proof, we will in fact have to
slightly strengthen our assumed estimates (1.3) and (1.6) as follows.
Extending to rp weighted estimates. First of all, we will need to extend (1.3) and (1.6) to rp -weighted
estimates, for 0 ≤ p < 2, of the type first introduced in [DR09]:
Z τ1
(p)
Z τ1
(p) (p−1) (p−1) (p)
E(τ ) + F(v) + χ 0 0
E (τ )dτ +0
E 0 (τ 0 )dτ 0 . E(τ0 ) + inhomogeneous terms,
−1
(1.7)
τ0 τ0
(p) (p)
Z τ1 (p−1)
Z τ1 (p)
Z τ1 ξ
F(v), E(τ ) + c ρ 0 0
E (τ )dτ + c 0 ρ 0
−1
0 0
E (τ )dτ + ≤ E(τ0 ) + A E 0 (τ 0 )dτ 0 + inhomogeneous terms.
−1
τ0 τ0 τ0
(1.8)
(p) (p−1) (−1−δ) (p) (p)
χ 0 p p−1 χ 0
In the above, E, E , etc., are r , respectively r , etc., weighted versions of E, E , etc, while E, F satisfy
(p) (p) (p) (p)
and we in fact assume that (1.8), just as (1.6), is the result of a pointwise coercive energy identity (1.4)
(p)
7
To obtain (1.7) and (1.8), given (1.3) and (1.6), it suffices to assume the existence of currents defined
only in the region r ≥ R̃ with suitable far away coercivity properties. For maximal generality, we will here
simply directly postulate the existence of such currents as an additional assumption. See already Section 3.5
for the precise formulation. This assumption can be shown to follow from suitable pointwise asymptotically
flat assumptions on the metric g0 and holds of course in all our examples of interest, where the currents can
easily be explicitly constructed. See Appendix B, and also [Mos16] for an even more general setting.
Extending to higher order estimates. Secondly, we will need, through suitable commutations, to
extend (1.3), (1.6), (1.7) and (1.8) to higher order statements. We will distinguish two energies
(p) (p)
E,
k
E
k
The energy Ek is the sum of the usual energy, with rp weights 0 ≤ p < 2, and with commutation
vector fields D e k up to order k, spanning the entire tangent space. The set will include the vector fields Ωi ,
i = 1, . . . , 3, which for large r correspond to the usual rotational vector fields. (Note that the latter will
be the only commutation vector fields which are r-weighted.) This is a fundamental energy with respect to
which initial data can be measured and with respect to which suitable Sobolev inequalities hold.
(p)
The energy E, k
on the other hand, denotes the precise energy flux of the current leading to (1.4) applied
moreover with a new set Dk of commutation vector fields, which will include T (and Ω1 , if this is assumed
to be Killing), but for which all non-Killing vectors are cut off to vanish in a suitable region of finite r. (See
already Sections 3.6.1–3.6.4 for the definition of these commutation vector fields and energies.) Nonetheless,
by our geometric assumptions and elliptic estimates, we will have that the energies
(p) (p)
E
k
∼ Ek (1.9)
(p) (p)
Z τ1 (p−1)
Z τ1 (−3−δ) (p)
Z τ1 ξ
ρ 0 ρ 0
F(v), E(τ
k
)+c 0
Ek (τ )dτ + c 0
k−1
0
E (τ )dτ ≤ E(τ
k
0
0) + A E 0 (τ 0 )dτ 0 + inhomogeneous terms, (1.11)
k−1
k
τ0 τ0 τ0
where again (1.11) derives from the pointwise coercivity properties of the energy identity (1.4) associated to
(p) (p) (p)
a current J,
k
whose flux terms are precisely given by E,
k
F. See already Section 3.6.8 for the precise form of
k
the estimates and inhomogeneous terms.
8
1.2 Assumptions on the nonlinearities g(ψ, x) and N (∂ψ, ψ, x)
We now turn to the nonlinear equation (1.1). In formulating assumptions on the allowed nonlinearities, we are
here largely motivated by geometric formulations of the Einstein equations, as in [DHRT21], which produce
“almost decoupled” equations similar to (1.1), which satisfy an appropriate form of the null condition. We
emphasise, however, that the resulting equations in such a formulation do not constitute a pure system of
wave equations in the form (1.1), and the quasilinear structure is mitigated through coupling with transport
equations. Thus, in this context, one should really only think of (1.1) as an indicative model equation. For
more specific discussion of the Einstein equations, see already Section 1.4.4.
In addressing the combined difficulties of the interaction of quasilinear terms, trapped null geodesics, and
merely polynomial decay at the level of model scalar equations of the form (1.1), the most natural simplest
setting is to require that the quasilinear term g(ψ, x) − g0 (x) be supported entirely in the region r ≤ R, and
to impose on the semilinear term N (∂ψ, ψ, x) a generalised version of the null condition [Kla86]. This also
allows us to use the exact null hypersurfaces of the background (M, g0 ) without additional complications,
making the null hierarchical structure of the estimates clearer. (In the case of the Einstein equations, in
our framework, these would be replaced by null hypersurfaces of the actual dynamic spacetime (M, g), for
instance associated to a double null gauge.) We note, however, that use of exact null hypersurfaces is in
no way fundamental; one can always replace these with suitable hyperboloidal hypersurfaces, for instance,
using the setup of [Mos16].
For maximum generality, rather than attempt a general algebraic definition of the null condition for
the semilinear terms N (∂ψ, ψ, x) of (1.1), we will formalise a “null condition assumption” in the form of
a time-translation translation invariant rp weighted estimate for the inhomogeneous terms of (1.10)–(1.11)
when applied to (1.1), restricted entirely to the asymptotic region r ≥ R.
To describe the condition, it will be convenient to introduce “master” energies
(p)
Z τ1
(0)
(p) (p−1) (−3−δ)
ρ 0
ρ
X
k
(τ ,
0 1τ ) := sup F(v)
k
+ sup E(τ
k
) + E 0 (τ ) dτ
Ek (τ ) + ρk−1 (1.12)
v τ0 ≤τ ≤τ1 τ0
finally X
k
where ρ is removed. Let us note that
(p) (p) (p) (p) (p)
ρ
X
k
. χX
k
.X
k
, X . χX
k−1 k
.
Specifically, our null condition assumption is that in any spacetime slab R(τ0 , τ1 ), the far away (i.e. r ≥ R)
contribution of the inhomogeneous terms arising from N (∂ψ, ψ, x) in the estimates (1.10), (1.11) can be
bounded by the expressions
(p)
q (0)
q q (p)
q (0)
q (p) (0+) (0+)
ρ ρ X (τ , τ ) ρ X (τ , τ ) X (τ , τ ), ρ
X
k
(τ 0 , τ1 ) X
k
(τ0 , τ1 ) + k
0 1 k
0 1 k
0 1 X
k
(τ0 , τ1 ) X (τ0 , τ1 )
k
(1.14)
for δ ≤ p ≤ 2 − δ and p = 0, respectively, and where all energies may in fact be restricted to the region
r ≥ 8R/9, where ρ = 1. (Note how the p = 0 weighted estimate is anomalous, in that the nonlinear terms
require the boundedness of a bulk term associated to the energy identity of a higher (p > 0) weight so as to
be estimated.) See already Section 4.7 for the precise formulation of the assumption.
The assumption that one can bound the relevant inhomogeneous terms arising from N (∂ψ, ψ, x) by (1.14)
isolates precisely that aspect of the usual null condition which is relevant for global existence in our method.
We will then show in Appendix C that our assumption indeed holds in particular for the usual nonlinearities
allowed by the classical null condition [Kla86] and includes also the class of semilinear terms N (∂ψ, ψ, x) on
Kerr studied previously in [Luk13].
9
1.3 The dyadically localised approach to global existence: sketch of the proof
of the main theorem
We now turn to the proof of the main theorem and the other novel aspect of the present work, namely the
replacement of a global bootstrap built on global decay-in-time estimates with dyadic iteration in consec-
utive spacetime slabs based entirely on dyadically localised, rp weighted—but time-translation invariant—
estimates.
We sketch the argument here. For details, see already Section 6. In our discussion below, we will
focus directly on our main case of interest, referred to later in the paper as case (iii), where the black box
inhomogeneous estimate (1.3) has nontrivial degeneration and indeed does not arise from a physical space
identity, and thus must be used in conjunction with the auxiliary identity (1.6). We emphasise, however,
that the dyadic approach described here is already novel in the case, referred to later in the paper as case (ii),
where the black box estimate (1.3) again has degeneration but is actually a consequence of a physical space
identity (as in Schwarzschild, see the discussion following (1.4)). Finally, in the case where the black box
estimate (1.3) derives from a physical space identity and is moreover non-degenerate (i.e. χ = 1 identically),
referred to in the paper as case (i), then dyadic iteration is in fact unnecessary, and the approach reduces to
a completely elementary time-translation invariant estimate. (This latter case applies for instance when g0 is
a small stationary perturbation of Minkowski space.) In the paper, we will provide self-contained treatments
of all three cases so the reader can compare with the more traditional approach.
(0)
q
0
√
(0)
+ C sup E(τ
k
) X
k
(τ 0 , τ ) L, (1.15)
τ0 ≤τ 0 ≤τ
The estimates are in fact contingent on an appropriate time-translation invariant bootstrap assumption on
the lower order energy
(p)
X (τ0 , τ ) . ε
k
(1.17)
q (0+)
q
(0+) (0+) (0+)
ρ ρ
for p = 0. The hierarchy (1.15)–(1.16) descends also to p = 0, but where X k
X (τ0 , τ ) and k−1
(τ0 , τ ) k X (τ0 , τ ) k
X (τ0 , τ )
replace the sum of the third to last and second to last terms of (1.15) and (1.16), respectively. (This
anomaly, referred to already immediately after (1.14), is a fundamental aspect of the estimates.) See already
Section 6.3.1. (p)
Estimate (1.15) arises from applying the physical space identity associated to the current J, k
which in the
linear case led to (1.11), directly to the k times commuted equation (1.1), i.e. it arises from the divergence
(p)
of J.
k
(The energies E
k
and F in (1.15) now in fact denote the exact fluxes arising from these currents; in
k
10
view of the bootstrap assumption (1.17), one can again show the equivalence (1.9) for solutions of (1.1).)
As discussed already in Section 1.1.2, it follows that (1.15) does not “lose derivatives”, as is clear from
examining the order of differentiability of the terms on the right hand side. We remark that the first boxed
term in (1.15) reflects the analogous term in (1.11). On the right hand side of (1.15), we recognise the
bound for the far away contribution of the nonlinear term (1.14) from our assumption capturing the null
condition, while the second boxed term in (1.15) is necessary to control the nonlinear terms on the set where
ρ degenerates, for there, this boxed term cannot be absorbed in the spacetime bulk term on the left hand
side. (The bad explicit dependence on the length L arises because one must take the supremum over τ for
the energy of the highest order terms there.)
Estimate (1.16), on the other hand, arises from applying the “black box” estimate (1.10) to the nonlinear
equation written in the form (1.5), i.e. now thought of simply as an inhomogeneous equation with respect
to the background g0 . The boxed term in (1.16), which is k’th order, reflects the loss of derivatives due to
the quasilinearity discussed already in Section 1.1.2.
The estimates can in principle close because, in view of our restrictions on the support of ξ, we have the
fundamental relation ξ (−1−δ) (−1−δ)
E 0 . χk−1
k−1
E 0 +k−2
E0 (1.18)
and thus we may bound the term Z τ ξ (0)
A E 0 (τ 0 )dτ 0 . χk−1
k−1
X (τ0 , τ ),
τ0
which is in turn bounded by the term on the left hand side of (1.16) for any p.
existence in an entire smaller slab of the form R(τ0 , τ0 + ), provided that some Ek . ε0 for sufficiently high k.
This is in fact true also for p = 0. (Note, that this “local” result already uses non-trivially the null condition
assumption of Section 1.2. This is because our foliation Σ(τ ) is outgoing null for r ≥ R. Equations (1.1)
not satisfying some version of the null condition can fail to yield solutions in R(τ0 , τ0 + ) for any > 0,
(p)
no matter how small the data are.) Moreover, we show that smallness of a suitably high Ek norm defines a
continuation criterion.
In view of the above, for global existence, it suffices to show that, under suitable assumptions at τ = τ0 ,
(p)
the quantity E,
k
for some p ∈ {0} ∪ [δ, 2 − δ] and suitably high k remains globally small in the slab. Examining
the hierarchy, for global existence in a slab R(τ0 , τ0 + L) of length L, it seems necessary to have at least
(0)
k
X (τ0 , τ ) . εL−1 . (1.19)
E(τ
k
0 ) . ε0 , E(τ0 ) . L−1 ε0 ,
k−2
(1.20)
which of course imply, for ε0 ε, both (1.17) (for p = 1) and (1.19) at τ = τ0 , an easy continuity argument,
with (1.17) and (1.19) taken as bootstrap assumptions, shows that if estimates (1.20) hold for τ = τ0 , then
the solution indeed exists globally in the entire slab R(τ0 , τ0 + L).
See already Section 6.3.2 for the details of the proof. Note that our above appeal to (1.17) and (1.19) is
in fact the only instance that a bootstrap assumption must be used in the proof of our main theorem.
11
τ = τ0 with the property that they are exactly recoverable at the top of the slab τ0 + L, with suitable
redefinition of ε0 and with L replaced by 2L.
Re-examining the hierarchy (1.15)–(1.16), we show that for an appropriate large constant α, and any
L ≥ 1, τ0 ≥ 1, then for ˆ0 sufficiently small, given for example the stronger estimates
(1) (2 − δ) (0) (1) (0)
E(τ
k
0 ) ≤ ε̂0 , E(τ0 ) ≤ ε̂0 ,
k−2
E(τ0 ) ≤ ε̂0 αL−1 ,
k−2
E(τ0 ) ≤ ε̂0 αL−1+δ ,
k−4
E(τ0 ) ≤ ε̂0 α2 L−2+δ , (1.21)
k−6
then (1.21) holds also at the final time τ0 + L where τ0 , L and ε̂0 are now however replaced by
1
τ0 + L, 2L, ε̂0 (1 + αL− 4 ), (1.22)
respectively. Note that the statement that the last three inequalities of (1.21) hold at τ0 + L with 2L in place
of L is a stronger smallness assumption than that assumed for these quantities at τ0 , signifying that these
quantities have in fact decayed, and relies on a pigeonhole argument as in [DR09], based on the fundamental
(p−1)
Ek 0 & E.
k
(1.23)
Briefly, the pigeonhole principle is applied as follows: The integral of the left hand side of (1.23) for p = 1
and k replaced by k − 2 can be shown to be bounded . ε̂0 , whence we obtain the existence of a τ = τ 00 slice
(0)
E(τ 00 ) is bounded . ε̂0 L−1 , where the L−1 factor arises from the length of the
for which the right hand side k−2
interval. We can then propagate this to τ0 + L, and use that α 1 to arrange that we have the precise new
(0)
existence of a τ = τ 00 slice for which the right hand side k−4 E(τ 00 ) is bounded . ε̂0 L−1 , the L−1 factor again
(2 − δ)
E(τ 00 ) . ε̂0 L−1+δ . Again, we can then propagate this to τ0 + L, and use that α 1 to
that on this slice k−4
(1)
arrange that we have the precise new version of the fourth inequality of (1.21) at τ0 + L, i.e. for k−4
E(τ0 + L).
Finally, a similar argument, applying (1.23) for p = 1 and k − 6, yields the new version of the last inequality
of (1.21).
Note that to obtain the new versions of the first two inequalities of (1.21) at τ0 + L, which refer to
quantities that do not decay with respect to their initial values, it is important that the constant appearing
in the first term on the right hand side of (1.15) is indeed 1.
E(τ
k
E(τ0 ) ≤ ε0
0 ) +k−2
for sufficiently small ε0 . It follows that (1.21) is satisfied for τ0 = 1, L := L0 = 1 and appropriate ε̂0 .
Examining closely (1.21) and (1.22), we see that we have now obtained a series of estimates which, in
addition to being sufficient to obtain existence in R(τ0 , τ0 +L), are moreover preserved under dyadic iteration.
We simply iterate the above statement on consecutive spacetime slabs of dyadic time-length Li = 2i , defining
−1
τi+1 = τi + Li = 2i , and ε̂i+1 = ε̂i (1 + αLi 4 ). Note that ε̂i+1 . ε̂0 . ε0 for all i.
This immediately yields global existence in R(τ0 , ∞) = ∪i≥0 R(τi , τi+1 ), the top order boundedness
statement (1) (2 − δ)
E(τ
k
E(τ ) . ε0
) +k−2 (1.24)
and the decay estimate
(0)
E(τi ) . ε0 τ −2+δ .
k−6
Thus, our result is a true orbital stability statement at highest order, in addition to yielding asymptotic
stability.
12
1.4 Discussion
We end with some additional discussion.
Other spacetimes (M, g0 ) satisfying our assumptions. We have already specifically mentioned the
examples where (M, g0 ) is Minkowski space, Schwarzschild and Kerr, but let us remark that our black box
assumption (1.3) holds in fact for the full subextremal Kerr–Newman family [Civ15]. We also note that
combining the use of an energy current construction similar to Appendix A with [WZ11], one can show
that (1.3) holds for arbitrary stationary perturbations (M, g0 ) of Schwarzschild, provided that they are close
to Schwarzschild in a suitably regular norm. Since we have already remarked that the physical space based
estimate (1.6) holds also for such spacetimes, in fact under much weaker regularity assumptions, this means
that our main theorem indeed applies in this case. Thus, for very slowly rotating |a| M Kerr spacetimes,
one sees that the special additional structure of Kerr (like the various manifestations of Carter separability,
the Killing tensor, etc.) has no significance for the problem, and the spacetimes can just be understood as
a very small stationary perturbation of Schwarzschild with a Killing horizon. It is only in the case |a| ∼ M
where true Kerr properties are of any relevance. A treatment of this case, reducing directly to the black box
estimate proven in [DRSR16], will appear elsewhere.
The axisymmetric case. In the case where we assume that the metric g0 admits an additional Killing
field Ω1 , one may also restrict to nonlinear equations of the form (1.1) which moreover preserve this symmetry,
i.e. with the property that if the data are preserved by Ω1 , then so is the solution. Let us call such equations
and solutions axisymmetric. Under the geometric assumptions of this paper, for axisymmetric equations (1.1)
and restricted to axisymmetric solutions, one may in fact easily derive the existence of the necessary auxiliary
physical space identity yielding (1.6), just as in Schwarzschild. This in particular applies to Kerr–Newman
in the full subextremal case. Thus, in view of [DRSR16, Civ15] which obtain (1.3), the main theorem of
the paper applies also in these settings. We leave the details as an exercise for the reader. Note that in this
case, if one prefers, one may quote the original [DR10] for (1.3) for axisymmetric solutions, in place of the
more general (1.3) obtained for all solutions in [DRSR16].
Potentials. Our method applies also if a suitably decaying T -independent potential V(x) is added to (1.1),
i.e. for
g(ψ,x) ψ − V(x)ψ = N (∂ψ, ψ, x),
provided that both the black box assumption (1.3) and the physical space based (1.6) hold for solutions of
the inhomogeneous version of the new linearised equation equation g0 ψ − V(x)ψ = F in place of (1.2). We
note that in the case where T is globally timelike and V decays suitably and is nonnegative, then, one can
13
always obtain also the analogue of (1.6). The same statement applies for any such potential in the slowly
rotating Kerr case |a| M .
Systems. For notational convenience, we have restricted here to scalar equations. The considerations
generalise readily to systems, for which a generalised null condition can again be formulated in terms of
bounds of the nonlinear terms by (1.14); this now includes many examples. See also the discussion of the
weak null condition in Section 1.4.5 below.
The obstacle problem. We have already remarked that, with a mild adaptation of the setup, one can also
consider the case where the boundary S is in fact timelike, imposing now also suitable boundary conditions
on S. This is for instance the setting for the classical obstacle problem. The analogue of (1.3) has indeed been
proven in the nontrapping case (with χ = 1 identically, i.e. without degeneration), but also, with suitable
degeneration functions χ, for many examples with nontrivial hyperbolic trapping (see for example [Laf22]).
We note again that the additional physical space estimate (1.6) can always be retrieved in this case. Thus,
this situation can also be incorporated in our framework.
Extremal black holes. Extremal black holes do not satisfy our assumptions on (M, g0 ), already because
the stationary vector field T will not be tangential to the boundary S, but more seriously, because χ must
degenerate at the black hole horizon [Sbi15], where T will not be timelike. Nonetheless, a version of (1.3)
has been proven in the extremal Reissner–Nordström case [Are11] (and in the extremal Kerr case, under the
assumption of axisymmetry [Are12]), with an additional hierarchy of degeneration associated to the horizon,
and nonlinear stability for semilinear equations has been proven [AAG20], where, however, an additional null
structural condition is required at the horizon, in full analogy with the situation at null infinity. It would
be interesting to reformulate this latter work in the dyadic setup used here. This is related to the nonlinear
stability problem of extremal black holes. See the discussion in [DHRT21] and Conjecture IV.2 therein.
The asymptotically AdS case. Finally, although our assumptions are modelled on the asymptotically
flat setting, one could also try to reformulate things in the asymptotically anti-de Sitter case, appropriate for
solutions of the Einstein equations with negative cosmological constant Λ < 0 (for a discussion of the Λ > 0
case see already Section 1.4.6 below). Here, one must also impose boundary conditions at infinity, since
infinity itself can be thought of as an asymptotic timelike boundary. We note that under reflecting boundary
conditions, there exist periodic (and thus non-decaying) solutions of the wave equation on pure AdS, while
general solutions on the Schwarzschild–anti de Sitter and Kerr–anti de Sitter case [HS13, HS14], again under
such boundary conditions, decay only inverse logarithmically. (This is due to stable photon orbits repeatedly
reflecting off of infinity only to return later having been repelled centrifugally by the black hole.) Based on
this lack of decay, pure AdS has in fact been proven to be nonlinearly unstable under reflecting boundary
conditions as a solution to various Einstein–matter systems, where the problem can be studied already under
spherical symmetry [Mos22]. Thus, if one hopes to show nonlinear stability for an equation of the type (1.1)
on asymptotically AdS backgrounds, it is more natural to consider dissipative boundary conditions like those
considered in [HLSW20]. Note that on Kerr–AdS, one expects to indeed satisfy the analogue of (1.6) with
the help of the Hawking–Reall Killing vector field, provided that the parameters satisfy the Hawking–Reall
bound. Thus, given a version of (1.3), nonlinear stability for a suitable class of equations (1.1) should be
tractable for all such Kerr–AdS parameters.
14
1
to pointwise decay t− 2 − 2 , considerably weaker than the t−1− “improved decay” which is often invoked for
global existence and stability for (1.1). Proving sharper decay is of course an extremely important problem in
itself (see [AAG16, Hin22, AAG21, Keh22]) but is not necessary (or even particularly helpful) for nonlinear
stability.
Let us also point out that for problems with gauge invariance, like the Einstein equations, dyadic local-
isation also provides a convenient opportunity to refresh the gauge. This suggests an alternative approach
to the global teleological normalisations of gauge done in [DHR, DHRT21].
1.4.4 Applications to the Teukolsky equation and derived equations and to the nonlinear
stability of black holes
As discussed in Section 1.2, one motivation for the precise assumptions on the nonlinearities of (1.1) made
here is viewing these as a model for understanding the Einstein equations, when the latter are considered
under suitable geometrically defined gauges. Such gauges were first exploited analytically in the monumental
proof of the nonlinear stability of Minkowski space [CK93].
The geometric-gauge approach to black hole stability was taken up in [DHR], where the problem was
studied in double null gauge, and the full linear stability of Schwarzschild was first proven. (See [Chr09] for
an introduction to double null gauge, including a discussion of the characteristic initial value problem and
a derivation of all equations.) The setup is of course intimately connected to the Newman–Penrose formal-
ism [NP62]. In the system resulting from linearising the reduced Einstein equations in double null gauge
around Schwarzschild, the gauge invariant quantities are determined by the extremal curvature components
α and α, which each satisfy the Teukolsky equation, first derived in this setting by [BP73] (and generalised to
Kerr in [Teu73]). The transport equations satisfied by the residual gauge dependent quantities, on the other
hand, are always already linearly coupled to the gauge invariant ones. To analyse first the gauge invariant
quantities, [DHR] introduced a pair of novel physical space quantities P and P , related to α and α by sec-
ond order weighted null differential operators, but satisfying the more tractable Regge–Wheeler equation (an
equation first derived in a different context in [RW57]), which, unlike the Teukolsky equation, could be under-
stood by the exact same methods as the linear wave equation. In particular, an analogue of (1.3) was proven,
via a physical space identity, for P and P , and this led to boundedness and decay through the hierarchical
structure of the system, first for the quantities P and P themselves, then for the original gauge-independent
quantities α and α, and then, after teleological normalisation of the double null gauge, of all residual gauge
dependent quantities as well. (For a complete scattering theory for this system, see [Mas22a, Mas22b]. For
other approaches to the gauge in Schwarzschild under linear theory, see [Ben22] and the references discussed
in Section 1.4.5 below.) The origin of this relation between Teukolsky and Regge–Wheeler goes back to the
fixed frequency transformation theory of [Cha75]. On the basis of this linear theory, the full nonlinear sta-
bility of the Schwarzschild family, without symmetry, was proven in our [DHRT21]. (For previous nonlinear
results for Schwarzschild under symmetry assumptions, see [Chr87, DR05] in the case of the Einstein-scalar
field system under spherical symmetry, and [Hol10, KS20] for vacuum, the former in the higher dimensional
case under biaxial Bianchi symmetry, and the latter under polarised axisymmetry, reducing to 2 + 1 di-
mensions, the first such result beyond 1 + 1 dimensional reductions.) Note that since Schwarzschild is a
co-dimension 3 subfamily of Kerr, nonlinear asymptotic stability of Schwarzschild refers to constructing the
full (teleologically defined) codimension 3 set of initial data which asymptote to Schwarzschild in the future.
A generalisation of the quantities P and P of [DHR] to the slowly rotating Kerr case |a| M was given
independently by [DHR19, Ma20]. The equations satisfied by these quantities are now however (weakly)
coupled to the quantities satisfying the Teukolsky equation. The works [DHR19, Ma20] both show an
analogue of (1.3) for this system, not via a physical space identity however but based on the framework for
frequency localised estimates on Kerr introduced in [DR10]. These results were followed by full linear stability
statements for the gauge dependent quantities in various gauges [ABBM19, HHV21] (the work [HHV21]
considers in fact harmonic gauge, cf. the discussion in Section 1.4.5). The estimates of [DHR19, Ma20] have
recently been reformulated by [GKS22] in the language of the higher order physical space commutation by
the Carter tensor first introduced in [AB15]. The work [GKS22] then uses this, among other ingredients, in
the context of the formalism [GKS20] to obtain nonlinear stability results for the very slowly rotating Kerr
case |a| M , culminating an impressive series of preprints. For generalisations to the Einstein–Maxwell
system see [Gio20, Gio21, Ape22].
15
The full subextremal Kerr case |a| < M is more subtle, as it cannot be understood as a small perturbation
of Schwarzschild. Remarkably, however, an analogue of (1.3) for this system has been obtained in the full
subextremal case [SRTdC20], using the already highly non-trivial mode stability results [Whi89, SR15,
TdC20] (see also [AMPW17]), but, also, non-trivial new high frequency structure with no apparent precise
analogue in the context of the wave equation. To exploit frequency analysis, the proof of [SRTdC20] uses a
version of the frequency localisation framework of [DRSR16].
In view of the method introduced in the present paper, it should be clear that there is absolutely nothing to
fear in these type of frequency localisations for nonlinear applications, provided that they are indeed used to
prove a spacetime localised statement which can be expressed in the form (1.3). The results of [DHR19, Ma20]
and [SRTdC20] can thus in principle be used directly for the nonlinear problem, in fact, as “black box” results.
Thus, in our view, given the breakthrough of [SRTdC20], not only is there no real conceptual difficulty in
extending the nonlinear Schwarzschild analysis of [DHRT21] to the full subextremal case |a| < M of Kerr,
but the technical complications are not as severe as one might naively have thought.
16
the time-slab localised point of view of the two works, the reason for the difference is clear: Since estimates
in [Mav21] have been reduced to a fixed time scale L, the role of (1.6) is essentially provided by Cauchy
stability, a completely soft statement.
From this point of view, the difference between the Schwarzschild and Kerr case on the one hand and
their de Sitter analogues on the other is more than simply one between polynomial and exponential decay,
but is one between dyadically localised and truly local. In this approach, it is this fundamentally local nature
of the analysis in the de Sitter case that renders the nonlinear aspects of stability problems on such de Sitter
backgrounds to be essentially soft.
1.6 Acknowledgements
MD acknowledges support through NSF grant DMS-2005464. GH acknowledges support by the Alexander
von Humboldt Foundation in the framework of the Alexander von Humboldt Professorship endowed by the
Federal Ministry of Education and Research and ERC Consolidator Grant 772249. IR acknowledges support
through NSF grants DMS-2005464 and a Simons Investigator Award. MT acknowledges support through
Royal Society Tata University Research Fellowship URF\R1\191409.
17
p
where f : [0, ∞) → [r0 , ∞) is a smooth function with f 0 (v) > 0 such that f (z) = r02 + z 2 near z = 0 and
f (z) = z for z ≥ 2r0 . Thus, in all cases r is a smooth positive function on M satisfying r ≥ r0 > 0. Let us
also fix a large R ≥ 20r0 .
We may also define ambient spherical coordinates in the usual way by the relation
In the case of no boundary, we will denote by Ω1 = ∂ϕ , Ω2 , and Ω3 the standard rotation vector fields
associated to the ambient spherical coordinates (x0 , r, ϑ, ϕ). These are globally regular vector fields.
In the case where the underlying structure is R4 , the above vector fields would only be regular in r > r0 .
In this case then, we will use the notation Ωi , i = 1, . . . , 3 for the above standard vector fields multiplied by
ω(r), for a smooth cutoff function satisfying ω(r) = 1 for r ≥ R/2, and ω(r) = 0 in a neighbourhood of r0 .
These are again globally regular vector fields.
2.2 The Lorentzian metric g0 , time orientation and the causality of the bound-
ary
We assume that g0 is a time oriented Lorentzian metric on M0 such that the coordinate vector field ∂x0 is
future directed timelike for r ≥ R/6.
In the case where M is a manifold with boundary S = {r = r0 }, we assume that the boundary S is
spacelike and the future directed normal points out of the spacetime.
T = ∂x0
with respect to the ambient coordinates (x0 , x1 , x2 , x3 ), we assume that T is Killing with respect to the
metric g0 . It follows by the previous assumptions that the ambient function r above is T -invariant and that
T is future directed timelike in the region r ≥ R/6. Let us further assume that g(T, T ) → −1 as r → ∞.
We will denote by φτ the 1-parameter group of diffeomorphisms generated by T .
In the case where S = ∅, it is natural to already assume that T is globally timelike, in view of the
Friedman instability [Mos18] and its evanescent analogue [Kei20] which applies in the marginal case, both
of which would be incompatible with assumptions we will make later on concerning the behaviour of waves.
In the case where S = 6 ∅, we notice that T cannot be globally timelike as T is tangential to and thus
spacelike on S = {r = r0 }. In this case we will need to assume the existence of a Killing horizon.
18
We assume that, for r0 ≥ R/2, Σ0 is transversal to the hypersurface {r = r0 }, and that Ω1 , Ω2 , Ω3 are
tangent to Σ0 ∩ {r ≥ R/2}. In particular the 2-dimensional space Tp (Σ0 ∩ {r ≤ R}) ∩ Tp (Σ0 ∩ {r ≥ R})
should coincide with the span of these vectors.
If S 6= ∅, we assume that Σ0 is transversal to the hypersurface S, and Σ0 ∩ S is diffeomorphic to the
2-sphere. We assume finally that the vector field Z of Section 2.4 is orthogonal to Σ0 ∩ H+ , while Ω1 , Ω2 ,
Ω3 are tangent to Σ0 ∩ H+ .
We assume that Σ0 separates M into two connected components and that Σ0 is a past Cauchy hyper-
surface for J + (Σ0 ).
Denoting by Σ(τ ) := (φτ )∗ (Σ0 ) we assume
J + (Σ0 ) = ∪τ ≥0 Σ(τ ),
and [
Σ(τ, v) := Σ(τ ) \ C ṽ .
ṽ>v
The spacetime region R(τ0 , τ1 , v) is a compact subset of spacetime with past boundary Σ(τ0 , v) and future
boundary S(τ0 , τ1 ) ∪ Σ(τ1 , v) ∪ C v .
19
2.5.4 Comparison of volume forms
We will assume that in the region r ≥ R, writing the volume form of (M, g) as
and, for Σ(τ ) ∩ {r ≥ R}, with the choice L for the normal, as
then
υ ∼ υ̃ ∼ υ̃˜ ∼ 1.
With this assumption, it follows by the coarea formula and compactness that globally, the volume form
of (M, g0 ) is related to the volume form of Σ(τ )
dVM ∼ dτ dVΣ(τ ) ,
2.6.2 / 2
The notation |∇ψ|
We define the notation X
/ 2 :=
|∇ψ| r−2 |Ωi ψ|2 .
i
We note that under our assumptions, the above expression is comparable to the induced gradient squared
on the spacelike spheres Σ(τ ) ∩ {r = r0 } in the region r ≥ R/2.
In view, however, of our spanning assumptions, in both the case S = ∅ and S =
6 ∅, the expression
will always be a translation invariant coercive expression on first derivatives of ψ, similar (by compactness)
to any other such coercive expression in r ≤ R, while in the region r ≥ R it will have the right weights in r
so as to be a suitable reference point for natural energy expressions.
20
2.6.3 The one-forms %L , %L , %1 , %2 and %3
We will need to introduce a set of weighted spanning 1-forms on the sphere so as to properly measure
weighted boundedness of forms. Define one forms %L , %L , and %1 , %2 , %3 on the far region M ∩ {r ≥ R} as
follows.
The region M ∩ {r ≥ R} can be written as the product manifold M ∩ {r ≥ R} = R × [R, ∞) × S 2 . Let
π : M ∩ {r ≥ R} → R × [R, ∞) denote the canonical projection. Let %L , %L be defined by %L = π ∗ %̃L and
%L = π ∗ %̃L , where %̃L , %̃L is the dual coframe of π∗ L, π∗ L on R × [R, ∞). It follows that
dψ = Lψ %L + Lψ %L + d
/ψ,
where d /ψ(L) = d
/ψ(L) = 0.
For each i = 1, 2, 3, there is a polar coordinate system (ϑi , ϕi ) on S 2 , associated to the corresponding
rotation vector field Ωi , with the property that, in the (x0 , r, ϑi , ϕi ) coordinate system for M ∩ {r ≥ R}, one
has ∂ϕi = Ωi . Such a (ϑi , ϕi ) coordinate system on S 2 is unique up to a choice of meridian. Define then, for
i = 1, 2, 3,
%1 = r sin ϑ1 dϕ1 , %2 = r sin ϑ2 dϕ2 , %3 = r sin ϑ3 dϕ3 .
There exist (non-unique, but universal) bounded functions ai j (ϑ, ϕ) for i, j = 1, 2, 3, such that, for any
smooth function ψ : M ∩ {r ≥ R} → R,
2.7.1 Minkowski
For the Minkowski case, we consider the underlying manifold to be R4 , i.e. without boundary, and we define
the metric to be the familiar expression
21
2.7.2 Schwarzschild
For the Schwarzschild case, given a real parameter M > 0, we may define r0 := (2 − δ1 )M for sufficiently
small 0 < δ1 M , denote x0 by t∗ , let r be defined by (1.1) and standard spherical coordinates (ϑ, ϕ) on
R3 by (2.3) and define the metric gM to be
We define the vector fields Ωi to be the standard Killing vector fields associated to spherical symmetry and
we define T = ∂t∗ to be the coordinate vector field with respect to the above coordinates.
Note that r = 2M is a Killing horizon with null generator T and positive surface gravity, and the
hypersurfaces r = r0 for r0 ∈ [r0 , 2M ) are indeed spacelike.
We may define the function t in the region r > 2M by t = t∗ − 2M log(r − 2M ). We note that in the
coordinates (t, r, θ, φ), the metric takes the familiar form
2.7.3 Kerr
To put Kerr in our preferred form, one uses a combination of Kerr star coordinates (based on Boyer–Lindquist
coordinates) and double null coordinates. We leave the details to the √ reader, but note the following.
Given subextremal parameters |a| < M , and defining r+ = M + M 2 − a2 , we set r0 = r+ − δ1 for a
small δ1 .
One may define the ambient differential structure so that the r of (2.1) will coincide with the Boyer–
Lindquist r of Appendix A in the region r ≤ R/3, while for r ≥ R/2 it will coincide with the coordinate r∗
of Appendix C.2. The coordinate v coincides with the double null v of Appendix C.2, and Σ(τ ) ∩ {r ≥ R}
will be a hypersurface of constant u, where again u is as defined in Appendix C.2.
Further, one may set things up so that Ω1 = ∂ϕ is the axisymmetric Killing field.
Let us note finally that extremal Kerr (corresponding to |a| = M ) cannot in fact be put into our preferred
form already because of our requirement that T be tangential to S and thus spacelike. Note also that the
Killing horizon of extremal Kerr has zero surface gravity, which would also contradict our assumptions of
Section 2.4.
22
r0 r ≥ r0 ; moreover if S =
6 ∅, then S = {r = r0 }
rKilling H+ = {r = rKilling } a Killing horizon with positive surface gravity;
span of T and Ω1 is timelike for r > rKilling
r1 parameter related to the vector field Y if S = 6 ∅
r1 + (r2 − r1 )/4 Z, Y , Ωi span the tangent space for r0 ≤ r1 + (r2 − r1 )/4
r1 + (r2 − r1 )/2 commutation vector fields D all Killing for r1 + (r2 − r1 )/2 ≤ r ≤ R/2
r2 ρ = 1, χ = 1 for r ≤ r2
R/6 T timelike for r ≥ R/6
R/4 ρ = 1, χ = 1 for r ≥ R/4
R/2 commutation vector fields D all Killing for r1 + (r2 − r1 )/2 ≤ r ≤ R/2 ;
g = g0 for r ≥ R/2
8R/9 the generalised null condition assumption concerns r ≥ 8R/9
R Σ(τ ) ∩ {r ≥ R} is null, C v ⊂ {r ≥ R}
R̃ parameter related to positivity properties of far-away currents
A.B
23
Let us thus define
χ : [r0 , ∞) → [0, 1] (3.2)
to be a function such that χ = 1 for r ≥ R/4 and r ≤ r2 .
The assumed estimate is:
(0)
Z τ
(0) (0)
Z (−1−δ)
τ
χ 0 0 0
sup F(v, τ0 , τ ), E(τ ) + cES (τ ) + c E (τ )dτ + c E 0 (τ 0 )dτ 0
−1
v:τ ≤τ (v) τ0 τ0
Z Z (3.3)
(0)
for some constants λ ≥ 1, C ≥ 1, 0 < c < 1, and where V0 is a fixed vector field and w0 a fixed function
satisfying X
|g0 (V0 , L)| . 1, |g0 (V0 , L)| . 1, |g0 (V0 , Ωi )|2 . r2 , |w0 | . r−1 . (3.4)
Here, the unprimed energies are defined by
Z (0)
ES (τ ) := |Lψ|2 + |∇ψ|
/ 2 + |Lψ|2 + ψ 2 , (3.6)
S(τ0 ,τ )
(0)
Z
F(τ0 , τ, v) := / 2 + r−2 ψ 2 ,
|Lψ|2 + |∇ψ| (3.7)
C v ∩R(τ0 ,τ )
(The 0 notation will be used in general to denote energies that naturally appear in bulk terms.) The
estimate (3.3) is to hold for all smooth ψ such that the right hand side of (3.3) is finite. In the above, we
already see the δ fixed in (3.1). For future reference let us also define the quantity
(−1−δ)
Z
E 0 (τ ) := r−1−δ (|Lψ|2 + |Lψ|2 + |∇ψ|
/ 2 ) + r−3−δ ψ 2 . (3.9)
Σ(τ )
(−1−δ) (−1−δ)
Note that if χ = 1 identically, then E 0 + −E1 0 = E 0 . We note that all expressions defined above are T -invariant.
We distinguish the constants C and λ in (3.3) to highlight the significance of the case λ = 1, when (3.3) is
derived from a suitable energy identity and the energies are replaced by exact fluxes. See already Section 3.4.1.
Again, the assumptions are motivated by our model cases of Minkowski space, Schwarzschild and sub-
extremal Kerr black hole exteriors. Note that (3.3) indeed holds in the case of Kerr in the full subextremal
range |a| < M [DRSR16]. In fact, in all of the above cases more precise estimates than (3.3) are available
with respect to what can actually be controlled by the right hand side; we shall not need to make use of this
here.
Remark 3.2.1. We in fact allow more generally on the right hand side of (3.3) a finite sum of terms of the
form |(V0µ ∂µ ψ + w0 ψ)F |, each with distinct V0 and w0 , but where all V0 and w0 satisfyR (3.4). For notational
convenience, we will always write a unique V0 , w0 , however. The inclusion of the term F 2 on the right hand
side of (3.3) is due to the fact that the estimate, as in [DRSR16], may be proven applying cutoff functions
to ψ, which result in such a term. In cases (i) and (ii) we shall see that this (3.3) holds without having to
include such a term. In case (iii), in practice, we can often restrict this term to be supported only in r ≤ R.
24
3.3 Physical space identities on a general Lorentzian metric
As explained in the introduction, for nonlinear applications to (1.1), it is most convenient when (3.3) is the
result of a physical space energy identity (1.4) for solutions ψ of (1.2). Thus, we will recall some general
properties of such identities here.
where
1 1 1
Tµν [g, ψ] = ∂µ ψ∂ν ψ − gµν g αβ ∂α ψ∂β ψ, X
πµν [g] = (LX g)µν = (∇µ Xν + ∇ν Xµ ),
2 2 2
and where ∗ : Λ3 M → Λ1 M denotes the Hodge star operator.
Let us note already that it is often more natural to parametrise choices of currents in a slightly different
way, as twisted currents [HW14], for instance, given a scalar function, one may define a twisted energy
momentum tensor T̃µν [g, ψ] according to (A.8) and then consider for instance currents of the form J˜µV [g, ψ] :=
T̃µν [g, ψ]V ν , etc. Such a current can always be rewritten as (3.10) for some w, q, $.
In particular, for solutions of the covariant wave equation g ψ = 0, one obtains a divergence relation
between the currents J and K which both depend only on the 1-jet of ψ. See [Chr00] for a discussion of the
classification of currents with this property. Note that $ does not contribute to the bulk current K, and
neither $ nor q contribute to H, which moreover is independent of the metric g.
The significance of identity (3.13) is that it can be integrated in a spacetime region bounded by homolo-
gous hypersurfaces to obtain a relation between boundary fluxes of J and a bulk integral of K. We give the
form of this relation below in the special case of the region R(τ0 , τ1 , v) ⊂ M.
where the normals and volume forms are with respect to the metric g0 according to the convention of
Section 2.5.4.
Identity (3.14) will be useful when it satisfies suitable positivity properties for its bulk and boundary terms.
25
3.4 Assumed unweighted first order physical space identities: cases (i)–(iii)
As discussed already in the introduction, we shall not in general require that (3.3) is the result of an integrated
divergence identity (3.14) associated to currents with a pointwise coercivity property. We shall, however,
require, in addition to assuming (3.3), the existence of currents generating an identity (3.14) with much
weaker non-negativity properties, properties which in particular are insensitive to the presence and structure
of possible trapping.
It is indeed useful to see first, however, how the existence of a physical space proof of (3.3) via an
identity of type (3.14) simplifies the considerations. Thus, we shall distinguish two simpler cases, to be
called case (i) and (ii), to be discussed in Sections 3.4.1 and 3.4.2 below, where indeed (3.3) is proven via an
identity (3.14), (with case (i) corresponding to the even simpler setting where there is no degeneration at all
in the coercivity).
The most general case, case (iii), which represents the main goal of this paper, will be discussed in
Section 3.4.3. We will introduce some helpful common notation in Section 3.4.4.
Finally, in Section 3.4.5, for the case S =
6 ∅, we will provide a family of currents with enhanced red-shift
control at H+ and in the black hole interior, parametrised by a parameter ς, which will be useful for obtaining
higher order estimates in Section 3.6.
JµV,w,q,$ [ψ]nµS 2 2 2 2
/
& (Lψ) + (Lψ) + |∇ψ| +ψ .
Here the normals are of course taken with respect to the metric g0 .
Let us note that the boundary coercivity statement on Σ(τ ) can only possibly hold if V is timelike on
Σ(τ ).
Defining Z Z
(0) (0)
µ
E(τ ) := V,w,q,$
Jµ [ψ]nΣ(τ ) , ES (τ ) := JµV,w,q,$ [ψ]nµS ,
Σ(τ ) S
(0)
Z (3.18)
F(v, τ0 , τ1 ) := JµV,w,q,$ [ψ]nµC ,
v
C v ∩R(τ0 ,τ1 )
26
it follows from (3.17) that
(0) (0) (0) (0) (0) (0)
From (3.15), it follows that in addition to the coercivity (3.17), we have the corresponding boundedness
JµV,w,q,$ [ψ]nµS 2 /
. (Lψ) + (Lψ) + |∇ψ| 2
+ψ , 2 2
Under the above assumptions, in the notation of Section 3.2 (recall now (3.9)), the identity (3.14) gives
rise to the estimate:
(0) (0)
Z τ1 (0)
Z
(−1−δ) (0)
We note that, in view of (3.19), (3.21) and the fact that we may express
H[ψ] = V µ ∂µ ψ + wψ,
this gives (3.3), for some λ ≥R1 and without degeneration, i.e. with χ = 1, and with V0 = V and w0 = w,
and where including the final F 2 term in (3.3) is here unnecessary. The point of expressing the estimate in
terms of the fraktur energies (3.18) is that (3.22) is a sharper statement than (3.3), corresponding to λ = 1,
which will be useful for us.
Let us note immediately that Minkowski space itself, but also suitably small stationary perturbations
of the Minkowski metric, satisfy the assumptions of this section (see Appendix B). More generally, given a
metric g0 as in Section 2 and a T -invariant quadruple (V, w, q, $) satisfying (3.15), whose energy currents
J V,w,q,$ [g0 , ψ], K V,w,q [g0 , ψ] satisfy the above coercivity properties (3.16), (3.17) and (3.20), it is clear that
J V,w,q,$ [g , ψ], K V,w,q [g , ψ] retain the coercivity properties (3.16), (3.17) and (3.20) for any stationary
perturbation g of g0 satisfying the assumptions of Section 2, sufficiently close to g, such that g = g in
r ≥ R. Thus, we see that when, as in the present section, estimate (3.3) is proven via (3.22), and there is no
degeneration, i.e. χ = 1, then the estimate (3.3) can immediately be inferred to be stable to suitably small
perturbations of the metric g0 .
Unfortunately, however, for most of our examples of spacetimes (M, g0 ) of interest, it turns out that the
assumptions of the present section cannot in fact hold. Specifically, if (M, g0 ) contains trapped null geodesics,
then one can show that (3.3) cannot hold with χ = 1 identically, and thus, no quadruple (V, w, q, $) as above
can exist satisfying (3.16); see [Sbi15]. In particular, the assumptions of this section do not encompass the
black hole cases of interest like Schwarzschild or Kerr.
27
where χ is the function (3.2) in Section 3.2.
We define the fraktur energies again by (3.18), and we note that again we have (3.19), and, in view
of (3.15), also (3.20), and thus (3.21). Under the above assumptions, the identity (3.14) gives rise to the
estimate Z τ Z τ
(0) (0) (0) (−1−δ)
χ 0 0 0
sup F(v, τ0 , τ ), E(τ ) + ES (τ ) + c E (τ )dτ + c E(τ 0 )dτ 0
−1
v:τ ≤τ (v) τ0 τ0
Z (3.24)
(0)
≤ E(τ0 ) + |H[ψ]F | .
R(τ0 ,τ )
We note again that, in view of (3.19) and (3.21), estimate (3.24) indeed implies (3.3), for some λ ≥ 1, with
V0 = V and w0 = w and where including the final F 2 term in (3.3) is here unnecessary. As with for (3.22)
R
of case (i), the point of expressing the estimate in terms of the fraktur energies (3.18) is that (3.22) is a
sharper statement than (3.3), corresponding to λ = 1, which will be useful for us.
We note that a current-defining quadruple (V, w, q, $) satisfying the properties of this section indeed exists
for the Schwarzschild metric and can be constructed from the considerations in [DR07a, MMTT10]. Note
that a pre-requisite for even the degenerate bulk coercivity property (3.23) is that any trapped null geodesics
be “unstable” in a suitable sense (cf. the Schwarzschild–AdS case with reflective boundary conditions at
infinity, where there exist stably trapped null geodesics [HS14]).
In the Kerr case for all |a| 6= 0, even though all trapped null geodesics are again unstable, and even
though estimate (3.3) is true (by [DRSR16]), one can show that no quadruple (V, w, q, $) can give rise to
currents satisfying the coercivity properties of this section (see [Ali09]). (For a higher order current defined
using second order operators which gives an analogue of the coercivity properties here in the |a| M case,
see however Andersson–Blue [AB15].)
One of the main motivations of this paper is to show that, from a suitable point of view, not only is
a purely physical space proof of (3.3) unnecessary for nonlinear applications, but such a proof would only
result in a minor and inessential simplification of the argument. We now turn to our main case of interest,
case (iii).
ξ = 0 in {χ 6= 1} ∪ {r ≤ r2 } ∪ {r ≥ R/4}, (3.26)
where χ is the function (3.2) in Section 3.2, appearing in the assumed estimate (3.3). (In nontrivial appli-
cations, ρ will vanish identically in a neighbourhood containing the set where χ 6= 1.) The supports of the
various functions is indicated in Figure 2. We note finally that we lose no generality in the present paper
28
χ
ρ
ξ
r0 r2 R/4
in always taking ρ, χ, ξ to be indicator functions of appropriate sets, but the sharpest estimates one could
prove would often involve degeneration on sets of positive codimension.
We assume the existence of a T -invariant quadruple (V, w, q, $) satisfying the boundedness proper-
ties (3.15), such that the associated current J V,w,q,$ [ψ] still satisfies (3.17), but where the bulk coercivity
assumption (3.23) on K V,w,q [ψ] is now further relaxed to
(−1−δ) (−1−δ)
and Z
ξ
0
E (τ ) :=
−1
ξ(r)ψ 2 . (3.29)
Σ(τ )
Again, we note that it is expressing things with respect to the fraktur energies which allows the constant to
be exactly 1 in the above estimate (3.28), and this fact will be useful for us.
The existence of currents J V,w,q,$ , K V,w,q satisfying the above assumptions can indeed be shown for
Kerr in the range |a| M (and in fact for general stationary suitably small perturbations of Schwarzschild
satisfying the assumptions of Section 2 and appropriate assumptions at infinity). See Appendix A. Note
that the estimate (3.28) is manifestly weaker than (3.3). The point, as discussed in the introduction, is that,
being derived from the (relaxed) coercivity properties of (3.27) applied to (3.13), estimate (3.28), or more
properly, the identity (3.13) itself, can be applied directly to (1.1). See already Section 4.3.1.
3.4.4 Summary of the unweighted assumptions for cases (i), (ii) and (iii)
So as to not have to always refer to separate formulas in the distinct cases (i), (ii) and (iii), we summarise
the assumptions in a way which can be subsequently interpreted for all cases simultaneously.
In case (i), we set ρ̃ = ρ = χ = 1 and A = Ã = 0.
In case (ii), we set χ to be the function (3.2) appearing in both (3.3) and (3.23), and we set ρ = χ, ρ̃ = 1
and A = Ã = 0.
29
Finally, in case (iii), we set χ to be the function (3.2) of Section 3.2 appearing in (3.3), we let ρ and ξ to
be the functions (3.25) and the constants A and à to be as in Section 3.4.3, and we set ρ̃ = ρ.
Our assumptions, applicable for all cases, are (a) that (3.3) holds and (b) that there exist T -invariant
(V, w, q, $), satisfying the bounds (3.15) such that, defining the currents J V,w,q,$ [ψ], K V,w,q [ψ] and H V,w [ψ]
by (3.10)–(3.12), we have the bulk coercivity property
K V,w,q [ψ] + Ãξ(r)ψ 2 & ρ(r)r−1−δ (Lψ)2 + (Lψ)2 + |∇ψ| / 2 + r−3 ψ 2 + ρ̃(r)r−3−δ ψ 2
(3.30)
JµV,w,q,$ [ψ]nµS 2 /
& (Lψ) + (Lψ) + |∇ψ| +ψ . 2 2 2
With these currents, we define the fraktur energies again by (3.18), and we again have (3.19), and, in
view of (3.15), also (3.20), and thus (3.21).
The energy identity (3.14) gives rise then to
(0) (0)
Z τ (0)
Z τ (−1−δ)
ρ 0 0
sup F(v, τ0 , τ ), E(τ ) + ES (τ ) + c E (τ )dτ 0 + c ρ̃ 0 0
E (τ )dτ 0
−1
v:τ ≤τ (v) τ0 τ0
Z τ Z (3.32)
(0) ξ
0 0 0
≤ E(τ0 ) + A E (τ )dτ +
−1
|H[ψ]F | ,
τ0 R(τ0 ,τ )
where we define Z
ρ̃ 0
E (τ ) :=
−1
ρ̃(r)r−3−δ ψ 2 .
Σ(τ )
We emphasise again that in cases (i) and (ii) the statement that (3.3) holds need not be taken as an
independent assumption, as (3.3) in fact follows from (3.32) with the above definitions in these two cases.
3.4.5 An enhanced redshift current and enhanced positivity in the black hole interior
In the case S =
6 ∅, for the purpose of higher order estimates to be considered in Section 3.6, we will need to
enhance the positivity near H+ . We have the following proposition.
Proposition 3.4.1. Under the assumptions of Section 3.4.4, there exists a constant c > 0 such that the
following holds.
Given arbitrary ς ≥ 1, there exist parameters r0 ≤ r00 (ς) < rKilling < r1 (ς) ≤ r1 , and a translation
0 0 0 0 0
invariant vector field Vς0 and one form qς0 , such that, defining J Vς ,w,qς ,$ [ψ], K Vς ,w,qς [ψ], H Vς ,w [ψ] as in
Section 3.4.4, then !
3
0 0 X
K Vς ,w,qς [ψ] ≥ c(Y ψ)2 + cς (Ωi ψ)2 + (T ψ)2 + ψ 2 (3.33)
i=1
in r00 (ς)
≤ r ≤ r1 (ς), and all properties of Section 3.4.4 hold for these currents with implicit constants
in (3.15), (3.30), (3.31), (3.20) which may be taken independent of ς. Finally, Vς0 = V , qς0 = q for r ≥ r1 .
Proof. This is clear by examining the proof of Theorem 7.1 of [DR13].
Using the above and the timelike character of ∇r in the “black hole interior” region r < rKilling , we may
further modify our current to obtain the following enhanced positivity in r0 ≤ r ≤ r1 (ς), in particular, all
the way up to S, at the expense of a lower order term. The resulting modified current will be used for higher
order estimates:
Proposition 3.4.2. Under the assumptions of Section 3.4.4, there exist constants C > 0, c > 0 such that
the following holds.
Given arbitrary ς ≥ 1, let Vς0 , qς0 be as given in Proposition 3.4.1. Then there exists a translation invariant
quadruple (Vς , wς , qς , $ς ) with Vς = Vς0 , wς = w, qς = q 0 , $ς = $ for r ≥ rKilling , and a positive function
30
λς (r), such that, defining J Vς ,wς ,qς ,$ς [ψ], K Vς ,wς ,qς [ψ], H Vς ,wς [ψ] as in Section 3.4.4, the following enhanced
positivity (modulo lower order terms):
3
!
X
Vς ,wς ,qς 2 2 2
K [ψ] ≥ cςλς (r) (Ωi ψ) + (T ψ) + |rKilling − r|(Y ψ)
i=1
+cλς (r)(Y ψ)2 − Cςλς (r)ψ 2 (3.34)
holds in r0 ≤ r ≤ rKilling .
Moreover, all boundedness and coercivity properties of the currents of Section 3.4.4 hold for these currents
as well with constants independent of ς in the region r ≥ rKilling , while in the region r < rKilling , the properties
of Section 3.4.4 again hold for these currents but with (3.30) replaced by (3.34), and with coercivity bounds
that now however depend in general on ς. In particular, we have
3
!
X
Vς ,wς
|H [ψ]| ≤ Cλς (r) |Y ψ| + |Ωi ψ| + |T ψ| + Cλς (r)ψ 2 . (3.35)
i=1
Proof. Given ς, let r00 (ς) be as given by Proposition 3.4.1. Let us define λς (r) to be a smooth positive
function such that λς (r) = 1 for r ≥ rKilling ,
dλς (r)
0≤ ≤ 2ςλ(r).
dr
We define now
For the first inequality, we are using the fact that ∇r is a smooth vector field which is future directed null
on H+ and future directed timelike in r < rKilling , by our assumptions from Section 2.4, as well as the fact
that !
3
1 µ 0ν X
Tµν [ψ] (∇ Vς + ∇ν Vς0 µ ) + w∇µ ψ∂µ ψ ≥ c (Y ψ)2 + (Ωi ψ)2 + (T ψ)2 , (3.37)
2 i=1
31
0 0
which follows because coercivity of the current K Vς ,w,qς asserted in (3.33) requires in particular coercivity of
its highest order terms. Note that the second inequality above clearly follows from the first wherever (3.36)
holds. On the other hand, wherever (3.36) does not hold, we again obtain the second inequality in view of
the definition of λς (r), since in this region, the enhanced positivity (3.33) applies, allowing us to put extra
ς factors on the second two terms of (3.37), as well as the bound ς(rKilling − r) ≤ 1.
On the other hand,
|K wς ,qς [ψ] − wς ∇µ ψ∂µ ψ| = |∇µ wς ψ∂µ ψ + ∇µ (qς )µ ψ 2 + 2ψ(qς )µ g µν ∂ν ψ|
3
!
dλς (r) X
≤C + λς (r) |ψ| |Y ψ| + |Ωi ψ| + |T ψ|
dr i=1
3
!
dλς (r) 2 dλς (r) X
+C + λς (r) ψ + C + Cλς (r) |ψ| |Y ψ| + |Ωi ψ| + |T ψ| ,
dr dr i=1
whence we deduce
1 Vςν
|K wς ,qς [ψ − wς ∇µ ψ∂µ ψ]| ≤ (K [ψ] + wς ∇µ ψ∂µ ψ) + Cςλς (r)ψ 2 .
2
On the other hand,
|H Vς ,wς [ψ]| = |Vςν ∂ν ψ| = |λς (r)Vς0 ν ψ + wς (r)ψ|
3
!
X
≤ Cλς (r) |Y ψ| + |Ωi ψ| + |T ψ| + Cλς (r)|ψ|,
i=1
giving (3.35). The statement (3.34) now follows since K Vς ,wς ,qς = K Vς + wς ∇µ ψ∂µ ψ + K wς ,qς − wς ∇µ ψ∂µ ψ
and the above bounds.
To give our precise assumption, let us first define Vfar , qfar , $far be as defined
in Appendix B.2. We assume then that for any δ ≤ p ≤ 2 − δ, there exists a T -invariant quadruple
(p) (p) (p)
(p) (p) (p)
◦ ◦ (p) (p)
◦ (p) (p)
(Vfar , wfar , qfar , $far ), with Vfar = Vfar + Ṽfar a vector field, wfar = wfar + w̃far a scalar function, qfar = qfar + q̃far
(p)
◦ (p)
a one form, and $far = $far + $̃far a two-form, defined on r ≥ R̃, satisfying
X
|g(Ṽfar , L)| . 1, |g(Ṽfar , L)| . 1, |g(Ṽfar , Ωi )|2 . 1,
|w̃far | . r−1 , |Lµ q̃farµ | . r−2 , |Lµ q̃farµ | . r−2 ,
|(∗ d$̃far )µ Lµ | . r−2 , |(∗ d$̃far )µ Lµ | . r−2 ,
(3.38)
| ∗ (%L ∧ $̃far )µ Lµ | . r−1 , | ∗ (%L ∧ $̃far )µ Lµ | . r−1 ,
| ∗ (%L ∧ $̃far )µ Lµ | . r−1 , | ∗ (%L ∧ $̃far )µ Lµ | . r−1 ,
| ∗ (%i ∧ $̃far )µ Lµ | . r−1 , | ∗ (%i ∧ $̃far )µ Lµ | . r−1 , i = 1, 2, 3 ,
(p) (p)
such that, defining the associated currents Jfar , Kfar , by (3.10), (3.11) these satisfy the weighted bulk
coercivity properties:
(p)
32
and the weighted boundary coercivity property
(p) p p
Jfarµ [ψ]nµΣ(τ ) & rp (r−1 L(rψ))2 + r 2 (Lψ)2 + |∇ψ|
/ 2 + r 2 −2 ψ 2 ,
(p)
(3.40)
Jfarµ [ψ]nµC & (Lψ)2 + rp |∇ψ|
/ 2 + rp−2 ψ 2 .
v
We note that the rp−1 (r−1 L(rψ))2 term is redundant in (3.39) as it can be estimated pointwise from
p−1
r (Lψ)2 and rp−3 ψ 2 . We retain it to compare with the boundary term (3.40) where it is necessary to
p p
retain explicitly the (r−1 L(rψ))2 term, as it is not controlled by the terms r 2 (Lψ)2 and r 2 −2 ψ 2 .
See Appendix B for the construction of such a current on Minkowski space and a broad class of space-
times with suitable asymptotic flatness assumptions at infinity, including Schwarzschild and Kerr in the full
subextremal range |a| < M .
Let us note immediately that given such currents satisfying the far away coercivity assumptions (3.39)
and (3.40), and given (V, w, q, $) as in Section 3.4.4, then by defining a suitable cutoff function ζ(r) with
ζ = 0 for r ≤ R̃ and ζ(r) = 1 for r ≥ R̃ + 1, and introducing a small fixed parameter e > 0, and defining the
T -invariant vector field, functions and forms
(p) (p)
(p) (p) (p) (p) (p) (p)
one sees immediately that the associated currents J, K satisfy the global (relaxed) weighted bulk coercivity
assumptions
(p)
as one can absorb the error terms in the region R̃ ≤ r ≤ R̃ + 1 arising from the cutoff by terms controlled
by the coercivity relations (3.30) and (3.31), provided e is suitably chosen.
From the bounds (3.38) and (3.43), we see that we have in fact
(p) p p
J µ [ψ]nµΣ(τ ) ∼ rp (r−1 L(rψ))2 + r 2 (Lψ)2 + |∇ψ|
/ 2 + ιr≤R (Lψ)2 + r 2 −2 ψ 2 ,
(p)
To state the resulting weighted versions of estimate (3.3) let us introduce some notation. For δ ≤ p ≤ 2−p
we define
Z
(p) (0) p p
E(τ ) := E(τ ) + rp (r−1 L(rψ))2 + r 2 (Lψ)2 + r 2 −2 ψ 2 , (3.45)
Σ(τ )∩{r≥R}
(p)
Z (0)
(−1−δ) (−1−δ)
We also define versions of (3.47) where the first term is replaced by χ E 0 (τ ) or ρ E 0 (τ ). We will call these
(p−1) (p−1)
χ 0
E (τ ) and ρ E 0 (τ ). Note that these latter two expressions do not control a zero’th order term in the region
r < R.
33
Note the following properties. For 2 − δ ≥ p ≥ δ, we have
(p) (p0 ) (p) (p0 ) (p−1) (p−1) (p−1) (0)
0 < δ ≤ p ≤ 2 − δ, the existence of a quadruple (Vfar , wfar , qfar , $far ) as above satisfying (3.38) and the far
away coercivity properties (3.39) and (3.40).
Then for all 0 < δ ≤ p ≤ 2 − δ, the following estimate holds for all τ0 ≤ τ :
(p)
Z τ Z τ
(p) (p) (p−1)
ρ 0 0
sup F(v, τ0 , τ ), E(τ ) + ES (τ0 , τ ) + c E (τ )dτ 0 + c ρ̃ 0 0
E (τ )dτ 0
−1
v:τ ≤τ (v) τ0 τ0
Z τ Z Z (3.52)
(p) ξ (p)
0 0 2
≤ E(τ0 ) + A E (τ ) +
−1
H[ψ]F + C F
τ0 R(τ0 ,τ ) R(τ0 ,τ )
where Ṽp is a fixed vector field and w̃p a fixed function satisfying
X
|g(Ṽp , L)| . 1, |g(Ṽp , L)| . 1, |g(Ṽp , Ωi )|2 . 1, |w̃p | . r−1 . (3.54)
Proof. The estimate (3.52) follows immediately from the energy identity (3.14) corresponding to the current
(p)
E 0 . −E1 0 .
−1
(0)
(0)
For consistency, we will in what follows often denote the quadruple (V, w, q, $) of Section 3.4.4 as V , w,
(0) (0) (0)
(0) (0)
Jς , Kς , Hς .
All boundedness and coercivity inequalities will continue to hold with constants independent of ς in the
(p)
region r ≥ rKilling , while in the region r < rKilling , the resulting Kς will satisfy the enhanced positivity
property (3.34), at the expense of lower order terms. (In the region r < rKilling , however, we emphasise the
dependence of the coercivity constants on ς.)
34
3.6 Higher order estimates
As is well known, in applications to nonlinear problems, one must be able to prove higher order estimates
in order for the estimates to close.
Let ζ(r) denote a smooth cutoff function such that ζ = 1 for r ≥ 3R/4 and ζ = 0 for r ≤ R/2. If S = ∅,
let ζ̂ = 0. Otherwise, let ζ̂(r) denote a smooth cutoff function such that ζ̂ = 1 for r ≤ r1 + (r2 − r1 )/4 and
ζ̂ = 0 for r ≥ r + 1 + (r2 − r1 )/2. If Ω1 is Killing, we define υ = 1, otherwise we set υ = 0. (More generally,
we may take υ = 0 if T coincides with the Killing generator and is timelike in r > rKilling .)
We define
Note that the vector fields (3.55) span the tangent space for r ≥ 3R/4 and, if S =6 ∅, in r ≤ r1 +(r2 −r1 )/4,
but not so in general for r1 + (r2 − r1 )/4 ≤ r ≤ 3R/4. It is also useful to have a collection of operators that
do. We thus define De k to denote commutation strings of operators from the collection
e 1 = L,
D e 2 = L,
D e 2+i = Ωi ,
D i = 1, . . . , 6, (3.57)
i.e. we define
e k1 D
e kψ = D
D e k2 · · · D
e k8 ψ. (3.58)
1 2 8
(Recall that the additional Ωi , i = 4, 5, 6 were introduced in the case S = ∅ to ensure the existence of a
convenient globally defined spanning set of vectors. In the case S =
6 ∅, we may understand the above formula
with Ω4 = Ω5 = Ω6 = 0.)
X (p)
1 X (p) X (p)
35
Proposition 3.6.1 ([DR13]). Let Y be the vector field of Section 2.6.4, and let H+ be a Killing horizon
with positive surface gravity as assumed in Section 2.4, such that the generator Z lies in the span of T and
Ω1 . Then, along H+ , we have
X
[Y k , g0 ]ψ = κk (ϑ, ϕ)(Y k+1 ψ) + Y k g0 ψ + αk (ϑ, φ)(Dk ψ),
|k|≤k+1,k3 <k+1
for a κk ≥ c > 0.
We have
(p) (p)
Corollary 3.6.2. Let H, Hς be as defined in Section 3.5. Then there exist constants c > 0, C > 0
(independent of ς) such that along H+ , we have
(p) X
H[Y k ψ][Y k , g0 ]ψ ≥ c(Y k+1 ψ)2 − C (Dk ψ)2 . (3.60)
|k|≤k+1,k3 6=k+1
Again, we emphasise that by our conventions of Section 3.1, the constants c and C on the right hand
side of (3.60) in general depend on k.
J[ψ]
k
:= κ(0, k)J[ψ] + κ(k, k)σ(k, k)Jς(k) [Dk ψ],
1≤|k|≤k
(p) (p) X (p)
(3.63)
K[ψ]
k
:= κ(0, k)K[ψ] + κ(k, k)σ(k, k)Kς(k) [Dk ψ],
1≤|k|≤k
H[ψ]
k
· Gk := κ0 (k)H[ψ]G0 + κ(k, k)σ(k, k)Hς(k) [Dk ψ]Gk , (3.64)
1≤|k|≤k
∇µ Jk µ [ψ] = K[ψ]
k
+ H[ψ]
k
· {[Dk , g0 ]ψ} + H[ψ]
k
· {Dk F }, (3.65)
36
which can again be integrated in region R(τ0 , τ1 , v) to yield
Z Z (p)
Z (p)
Z (p)
J[ψ]
k
· n + J[ψ] · n + J[ψ]
k
·n+ K[ψ]
k
Σ(τ1 ,v) S(τ0 ,τ1 ) C v (τ0 ,τ1 ) R(τ0 ,τ1 ,v)
Z (p)
Z (p)
Z (p)
= J[ψ]
k
·n− H[ψ]
k
· {[Dk , g0 ]ψ} − H[ψ]
k
· {Dk F }. (3.66)
Σ(τ0 ,v) R(τ0 ,τ1 ,v) R(τ0 ,τ1 ,v)
H[ψ]
k
· {[Dk , g0 ]ψ} = 0 (3.67)
in r1 + (r2 − r1 )/2 ≤ r ≤ R/2.
Finally, if
κ0 (k) ς(k) (3.68)
(p)
K[ψ]
k
≥ cκ0 (k)K[ψ] (3.69)
3
!
X X
+c κ(k, k)ςλς (r) (Ωi Dk ψ)2 + (T Dk ψ)2 + |rKilling − r|(Y Dk ψ)2 + κ(k, k)λς (r)(Y Dk ψ)2 ,
1≤|k|≤k i=1
where we have absorbed the lower order term in (3.34) with the wrong sign by the largeness of κ0 (k) =
(p)
κ(|k|, k)−1 κ(|k| − 1, k) and the positivity of K[ψ], and we have dropped the factor σ(k, k) using that σ ≥ 1.
We will always assume (3.68) in what follows so that (3.69) holds.
Defining Z Z
(p) (p) (p) (p)
E(τ
k
) := Jk µ [ψ]nµΣ(τ ) , E
k S
(τ0 , τ ) := Jk µ [ψ]nµS ,
Σ(τ ) S(τ0 ,τ )
(p)
Z (p)
(3.70)
F(v, τ0 , τ ) := Jk µ [ψ]nµC ,
k v
C v ∩R(τ0 ,τ )
E(τ
k
) ≥ 0, E
k S
(τ0 , τ ) ≥ 0, F(v, τ0 , τ ) ≥ 0,
k
+ |H[ψ]
k
· {[Dk , g0 ]ψ}|
R(τ0 ,τ1 )∩{r≥R/2}
Z (p)
+ |H[ψ]
k
· {Dk F }|, (3.71)
R(τ0 ,τ1 )
provided the terms on the right hand side of the above inequalities are suitably integrable.
37
Proposition 3.6.3. Let (M, g0 ) satisfy the assumptions of Section 2. Let ψ be a solution of the inhomoge-
neous equation (4.4) in R(τ0 , τ1 ) and let τ0 ≤ τ 0 ≤ τ1 . Then for all k ≥ 1 and for all rKilling < r−
0
< r0 <
00 00
r < r+ ≤ R,
Z X Z X X X
k 2
(D ψ) .
e (Dk ψ)2 + e k ψ)2 +
(D (De k F )2 ,
Σ(τ 0 )∩{r 0 ≤r≤r 00 } |k|≤k+1 0 ≤r≤r 00 }
Σ(τ 0 )∩{r− + 1≤|k|≤k+1,k1 +k2 ≥|k|−1 |k|≤1 |k|≤k−1
(3.72)
0
where here . depends on the choice of r− < r0 < r00 < r+
00
. We also have the estimate, for all r0 ≤ r1 ,
Z X Z X X
k 2
(D ψ) .
e (Dk ψ)2 + e k F )2 .
(D (3.73)
Σ(τ 0 )∩{r≥r 0 } |k|≤k+1 Σ(τ 0 )∩{r≥r 0 } |k|≤k+1 |k|≤k−1
Note that the analogous statements to (3.72), (3.73) with integration on R(τ0 , τ1 ) ∩ {r0 ≤ r ≤ r00 }, etc.,
follows immediately in view of the coarea formula.
00
In fact, even without assuming r+ ≤ R, we still have the spacetime elliptic estimate:
Z X Z X X X
(De k ψ)2 . (Dk ψ)2 + e k ψ)2 +
(D e k F )2 ,
(D
R(τ0 ,τ1 )∩{r 0 ≤r≤r 00 } |k|≤k+1 0 ≤r≤r 00 }
R(τ0 ,τ1 )∩{r− + 1≤|k|≤k+1,k1 +k2 ≥|k|−1 |k|≤1 |k|≤k−1
(3.74)
0
where again . depends on the choice of r− < r0 < r00 < r+
00
.
Proof. Estimate (3.72) is a standard elliptic estimate, using the fact that the span of T, Ω1 always contains
0
a timelike direction in the region r ≥ r− > rKilling . Estimate (3.73) then follows from the previous in view
k
of the fact that the D span the tangent space for r ≤ r1 and r ≥ R/2.
00
As we have remarked, in the case r+ ≤ R, estimate (3.74) follows from (3.72) by the coarea formula. It
can be obtained more generally, even if R ≤ r00 or R ≤ r+ 00
by a suitable integration by parts argument. We
sketch this here for the case k = 1. Let us consider the most difficult case where R < r00 < r+ 00
. One first
does the usual elliptic estimates on Σ(τ ) ∩ {r ≤ R}, then integrating over τ , to estimate the left hand side
of (3.74) integrated over R(τ0 , τ1 ) ∩ {r ≤ R} from the right hand side, but with an additional boundary term
on r = R. Squaring the inhomogeneous wave equation in r ≥ R and integrating by parts along the null cone
Σ(τ ) ∩ {r ≤ R}, one may again bound the left hand side of (3.74), integrated over R(τ0 , τ1 ) ∩ {R ≤ r ≤ r00 },
from the right hand side and an additional
R boundary term on r = R. These boundary terms can be arranged
to cancel modulo a term of the form R(τ0 ,τ1 )∩{r=R} T ψ ∆ψ,/ where ∆ / denotes the Laplacian on Σ(τ )∩{r = R},
and this term can in turn be related to a bulk integral which can again be controlled by the right hand side
of (3.74).
38
Proof. We first prove (3.79). Here we will use (3.59). We have that the left hand side of (3.79) is bounded
in r ≥ Rk by
(p) X (p)
H[ψ]
k
· {[Dk , g0 ]ψ}|k|≤k = κ(k, k)H[Dk ψ][Dk , g0 ]
|k|≤k
1 X (p) X (p)
−H[ψ]
k
· {[Dk , g0 ]ψ} = − κ(|k|, k)Hς(k) [Dk ψ] [Dk , g0 ]ψ − κ(k, k)Hς(k) [Y k ψ][Y k , g0 ]ψ
1≤|k|≤k,k3 <k
(3.81)
3
!
1
X X
≤ Cς(k) 2 λς(k) (r) κ(|k|, k) (T Dk ψ)2 + (Ωi Dk ψ)2 + (Dk ψ)2
1≤|k|≤k i=1
− 21
X
+ Cς(k) λς(k) (r) κ(|k| − 1, k)(Dk ψ)2 (3.82)
1≤|k|≤k+1
X
− cλς(k) (r)κ(k, k)(Y k+1 ψ)2 + Cλς(k) (r)|r − rKilling |κ(k, k) (Dk ψ)2 .
1≤|k|≤k+1
(p)
Here, we have applied (3.60) from Corollary 3.6.2, and the bounds on Hς(k) following from (3.35) of Propo-
sition 3.4.2. Note the dependence on ς(k) through λς(k) (r), and note that the terms in the commutation
identity not present on H+ appear with an extra (r − rKilling ) factor. (We emphasise again the conventions
from Section 3.1 that constants C, c, also depends on k!)
We now have that given ς = ς(k), from (3.69), it follows that in the region r0 ≤ r ≤ r1 (ς), provided
that (3.68) holds, we have
3
!
(p)
1 1
X X
ς(k) 2 λς(k) (r) k 2
κ(|k|, k) (T D ψ) + k 2
(Ωi D ψ) + (D ψ)k 2
≤ Cς − 2 (k)K[ψ].
k
1≤|k|≤k i=1
(p)
1 1
X
ς − 2 (k)λς(k) (r) κ(|k| − 1)(Dk ψ)2 ≤ Cς − 2 (k)K[ψ].
k
1≤|k|≤k+1
Finally, again provided (3.68) holds, and provided we further restrict r1 (ς) such that 0 < r1 (ς)−rKilling ≤
ς −1 , we have
X (p)
|k|≤k+1
39
1
Now we fix ς = ς(k) to be sufficiently large such that Cς − 2 (k) ≤ 13 . This yields (3.75).
Since ς(k) is now fixed, as well as r1 (k) := r1 (ς(k)), according to our conventions, the dependence on
these parameters may now be absorbed into the constants C, c, etc.
To prove (3.76), we estimate
X Z (p)
|H[ψ]
k
· {[Dk , g0 ]ψ}|
|k|≤k R(τ0 ,τ1 )∩{r1 (k)≤r≤r2 }
Z X
≤C e k ψ)2
κ(|k| − 1, k)(D
R(τ0 ,τ1 )∩{r1 (k)≤r≤r2 } |k|≤k+1
Z X
≤C κ(|k| − 1, k)(Dk ψ)2
R(τ0 ,τ1 )∩{rKilling ≤r≤r2 } 1≤|k|≤k+1,k +k ≥|k|−1
1 2
X X
+ e k ψ)2 + C
(D e k F )2
(D
|k|≤1 |k|≤k−1
Z (p) (p) X
−1
≤C σ12 (k)K[ψ]
k
+ Cκ −1 (0, k)K[ψ]
k
+C e k F )2 .
(D
R(τ0 ,τ1 )∩{rKilling ≤r≤r2 } |k|≤k−1
(We emphasise again the conventions from Section 3.1 that constants C also depends on k.) We have
0
used the spacetime version of estimate (3.72) of Proposition 3.6.3 in the above (i.e. (3.74)), with r− :=
1 00 0 00
rKilling + 2 (r1 (k) − rKilling ), r+ := r1 + 3(r2 − r1 )/4, r = r1 (k), r = r1 + (r2 − r1 )/2, where we also use that
the integrand on the left hand side is nonzero only in r0 ≤ r ≤ r00 .
Requiring now σ12 (k) 1 to be sufficiently large and κ(0, k) = κ0 (k) 1 to be sufficiently large, we
obtain (3.76).
The proof of (3.78) is analogous to (3.76), and also constrains σ12 (k), κ0 (k) to be sufficiently large. This
finally fixes σ12 (k) and κ0 (k). Note here we use (3.74) with r00 = Rk and r+ 00
= 9Rk /8 and in general we
may have 9Rk /8 ≥ R.
Identity (3.77) follows immediately from the definition of the commutation vector fields D.
In what follows, we shall consider ς, σ and κ fixed so as to satisfy the above Proposition. Thus, from
now on, dependence of constants on ς, σ and κ will be absorbed in the C and . notation. We emphasise
again that, according to our conventions from Section 3.1, all our constants c, C will in general depend on
k.
We have the following immediate corollary of Proposition 3.6.4:
Corollary 3.6.5. Let k ≥ 1 and let ψ be a solution of the inhomogeneous equation (4.4) in R(τ0 , τ1 ). Then
we have the global (one-sided) bound:
Z (p)
Z (p)
k
− H[ψ]
k
· {[D , g0 ]ψ} + |H[ψ]
k
· {[Dk , g0 ]ψ}|
R(τ0 ,τ1 )∩{r≤r2 } R(τ0 ,τ1 )∩{r≥r2 }
Z
1 (p)
X (3.84)
≤ K[ψ] + C |De k F |2 .
R(τ0 ,τ1 )∩{r≤r2 }∩{r≥R/4} 2
k
|k|≤k−1
Proof. Note that the presence of 21 instead of 13 is due to the fact that both estimates (3.78) and (3.79)
borrow from the bulk of the region r ≥ Rk . Note that restricted to the region r ≥ R, if desired, we may of
course replace the D
e commutation on the last term on the right hand side by D.
40
In general, our fundamental energies without degeneration functions χ or ρ will be defined with D e k as
commutation vector fields, i.e. they will contain all derivatives at the relevant order. Specifically, we define
first the unweighted energies:
X Z (0)
|k|≤k Σ(τ )
(0) X Z
Ek S (τ0 , τ ) := e k ψ)2 + |∇ψ|
(LD / 2 + (LD
e k ψ)2 + (D
e k ψ)2 , (3.86)
|k|≤k S(τ0 ,τ )
(0) X Z
F(v, τ0 , τ ) := e k ψ)2 + (∇
(LD /De k ψ)2 + r−2 (D
e k ψ)2 , (3.87)
k
|k|≤k C v ∩R(τ0 ,τ )
(−1−δ) X Z
0
Ek (τ ) := r−1−δ (LD
e k ψ)2 + (LD
e k ψ)2 + |∇
/De k ψ|2 + r−3−δ (D
e k ψ)2 . (3.88)
|k|≤k Σ(τ )
We now define (in analogy with (3.45)–(3.47)) the higher order p-weighted energies for δ ≤ p ≤ 2 − δ:
X Z
e k ψ))2 + r p2 (LD
e k ψ)2 + r p2 −2 (D
(p) (0)
E(τ
k
) := E(τ
k
)+ rp (r−1 L(rD e k ψ)2 , (3.89)
|k|≤k Σ(τ )∩{r≥R}
(p) (0) X Z
F(v,
k
τ0 , τ ) := F(v)
k
+ rp |∇
/De k ψ|2 + rp−2 (D
e k ψ)2 , (3.90)
|k|≤k C v ∩R(τ0 ,τ )
(p−1) (−1−δ) X Z
Ek 0 (τ ) := Ek 0 (τ ) + rp−1 (r−1 L(rD
e k ψ))2 + (LD
e k ψ)2 + |∇
/De k ψ|2 + rp−3 (D
e k ψ)2 . (3.91)
|k|≤k Σ(τ )
Ek & E,
k
F
k
&F
k
for p ≥ p0 ≥ δ or p0 = 0, Ek 0 & Ek for p ≥ 1 + δ, Ek 0 & Ek for p ≥ 1, (3.92)
(−1−δ) X Z
ρ 0
r−1−δ ρ(r) (LDk ψ)2 + (LDk ψ)2 + |∇D
/ k ψ|2 ,
Ek (τ ) := (3.94)
|k|≤k Σ(τ )
(−3−δ) X Z
ρ̃ 0
E (τ ) :=
k−1
ρ̃(r)r−3−δ (Dk ψ)2 (3.95)
|k|≤k Σ(τ )
and then
(p−1) (−1−δ) X Z
χ 0 χ 0
Ek (τ ) := Ek (τ ) + rp−1 (r−1 L(rD
e k ψ))2 + (LD
e k ψ)2 + |∇
/De k ψ|2 + rp−3 (D
e k ψ)2 , (3.96)
|k|≤k Σ(τ )∩{r≥R}
(p−1) (−1−δ) X Z
ρ 0
Ek (τ ) = ρ Ek 0 (τ ) + rp−1 (r−1 L(rD
e k ψ))2 + (LD
e k ψ)2 + |∇
/De k ψ|2 + rp−3 (D
e k ψ)2 . (3.97)
|k|≤k Σ(τ )∩{r≥R}
(Note that in the integrals over r ≥ R, it would not matter whether we use D
e or D as these would here
define directly comparable energies.)
In analogy with (3.29) we define
ξ X Z
0
E (τ ) :=
k−1
ξ(r)(Dk ψ)2 . (3.98)
|k|≤k Σ(τ )
41
We note the fundamental relation ξ (−1−δ) (−1−δ)
k−1
E0
E 0 +k−2
E 0 . χk−1 (3.99)
which follows from our constraints on the support of ξ and the degeneration of χ.
Proposition 3.6.6. For a general smooth function ψ, we have the following relations between energies
(p) (p)
E(τ
k
) . E(τ
k
), (3.100)
(p) (0) (p) (p) (p)
Ek S
(τ ) = E
k S
(τ ) ∼ Ek S (τ ), F(v, τ0 , τ ) ∼ F(v,
k
τ0 , τ ) (3.101)
k
Z (p) X Z τ1 Z τ1 (p−1) (−3−δ)
ρ 0 0 ρ̃ 0 0
K[ψ]
k
+ Ãξ(r)(Dk ψ)2 & Ek (τ )dτ 0 + E (τ )dτ 0 ,
k−1
2−δ ≥p≥δ
R(τ0 ,τ1 ) |k|≤k τ0 τ0
Z (p) X Z τ1 (−1−δ)
Z τ1 (−3−δ)
(3.102)
k 2 ρ 0 0 0 ρ̃ 0 0 0
K[ψ]
k
+ Ãξ(r)(D ψ) & Ek (τ )dτ + k−1
E (τ )dτ , p = 0.
R(τ0 ,τ1 ) |k|≤k τ0 τ0
E(τ ) . E(τ ) + e k F )2 .
(D (3.103)
k k
Σ(τ )∩{r≤R} |k|≤k−1
Proof. The relations (3.100)–(3.102) follow immediately from our coercivity and boundedness assumptions
on the currents while the inequalities (3.103)–(3.107) follow easily from the elliptic estimate (3.73) of Propo-
sition 3.6.3.
We note the following immediate corollary of the above proposition:
Corollary 3.6.7. Let ψ be a solution of g0 ψ = 0 in R(τ0 , τ1 ). Then calligraphic and fraktur energies on
Σ(τ ) are equivalent
(p) (p)
E(τ
k
) ∼ E(τ
k
).
If χ = 1 and ρ̃ = 1 identically (as in case (i)), then
(p−1) (p−1) (−3−δ) (−1−δ) (−1−δ) (−3−δ)
Ek 0 (τ ) ∼ χ Ek 0 (τ ) + ρ̃k−1
E 0 (τ ), Ek 0 (τ ) ∼ χ Ek 0 (τ ) + ρ̃k−1
E 0 (τ ).
E 0 (τ ) . ρ Ek 0 (τ ) + ρ̃k−1
k−1
E 0 (τ ), E 0 (τ ) . ρ Ek 0 (τ ) + ρ̃k−1
k−1
E 0 (τ ).
42
3.6.8 The final higher order estimates
We conclude with our higher order version of Proposition 3.5.1.
Proposition 3.6.8. Under the assumptions of Proposition 3.5.1, let us assume the additional assumptions
of Section 3.6.2.
For all k ≥ 1. Then for all 0 < δ ≤ p ≤ 2 − δ and for all τ0 ≤ τ ≤ τ1 we have the following statement.
Let ψ be a solution of the inhomogeneous equation (4.4) in R(τ0 , τ1 ). Then
(p)
Z τ
(p)
Z τ
(0) (p−1) (−3−δ)
ρ 0 0 0 ρ̃ 0 0
sup F(v, τ0 , τ ), E(τ
k
) + E
k S
(τ0 , τ ) + c E
k
(τ )dτ + c E (τ )dτ 0
k−1
v:τ ≤τ (v) τ0 τ0
(p)
Z τ ξ
Z (p)
Z X
0 0 k
≤ E(τ
k
0) + A E (τ ) +
k−1
H[ψ]
k
· {D F } + C (Dk F )2
τ0 R(τ0 ,τ ) R(τ0 ,τ ) |k|≤k
Z X
+C e k F )2
(D (3.108)
R(τ0 ,τ )∩{r≤R} |k|≤k−1
where we recall that υ = 1 only if Ω1 is Killing. It follows that the left hand side of (3.108) is bounded by
the right hand side of (3.109).
Now apply the black box estimate (3.53) to Dk ψ, and note that we can bound the terms arising from
Rτ (p−1) (−3−δ)
[g0 , Dk ]ψ by c τ0 ρ Ek 0 (τ 0 )+k−1
E 0 (τ 0 )dτ 0 and the spacetime term involving De k F on the right hand side of (3.109).
We thus have that (3.110) holds with Dk ψ in place of T k ψ.
(p) (p)
43
Finally, by (3.73) of Proposition 3.6.3, we may estimate
Z τ (p−1)
Z τ (p−1)
(−3−δ)
Z X
0 0 0 ρ 0 0 0 0 0 e k F |2 .
E (τ )dτ .
k−1
E (τ ) +k−1
k−1
E (τ ) dτ + |D
τ0 τ0 R(τ0 ,τ )∩{r≤R} |k|≤k−1
Remark 3.6.9. Note that for convenience we have used (3.108) to obtain (3.109). This was because our
precise commutation assumptions were stated with respect to the global currents. If one does not assume
estimate (3.32), and thus one does not have (3.108), one may still obtain (3.109) under the assumption
(p) (p)
of asymptotic flatness of Sections 3.5 and 3.6.2, where we insert simply the far-away currents Jfar , Kfar
in (3.59). (We also use that, from the assumptions of Section 2, we can still define currents as in Section 3.4.5
giving enhanced positivity near H+ and in the black hole interior.) Thus, in particular, estimate (3.109) holds
in the Kerr case in the full subextremal range |a| < M .
Proof. We note that the left hand side of (3.111) is in fact bounded by the energy restricted to Σ ∩ {r ≤ R},
which is why we may choose either of the two quantities on the right hand side (3.111).
Weighted Sobolev inequalities will also hold globally on Σ(τ ), and in practice these are used to estimate
nonlinearities in the region near infinity. Because this use is incorporated in our assumption on the null
condition (see already Section 4.7), we shall not need to state such inequalities, although, in practice, they
will appear in the context of verifying the assumptions of Section 4.7. See already Appendix C.
Finally, we have the following easy interpolation result:
Proposition 3.6.11. For δ as fixed in (3.1), one has the following interpolation inequalities
(1)
1−δ
(1 − δ)
δ
(2 − δ)
E(τ
k
). E(τ
k
) E(τ
k
) , (3.112)
1−δ
1+δ 2δ
1+δ
(δ − 1) (−1−δ) (0)
Ek 0 (τ ) . Ek 0 (τ ) Ek 0 (τ ) , (3.113)
s s
Z τ0 + 21 (0)
Z τ0 + 12 (0)
Z τ0 + 12 (0)
E(τ ) .
k−1
E(τ
k
) E(τ ).
k−2
(3.114)
τ0 τ0 τ0
Proof. The proof of these inequalities is standard and is left to the reader.
44
4.1 The class of equations
We will consider solutions ψ to quasilinear equations of the form
where
g : R × M → T ∗ M ⊗ T ∗ M, N : R × M → TM ⊗ TM (4.2)
∗ ∗
are such that π ◦ g(ψ, x) = x, π ◦ N (ψ, x) = x where π : T M ⊗ T M → M, π : T M ⊗ T M → M denote the
canonical projections, and such that g(0, x) = g0 and g(·, x) = g0 for r(x) ≥ R/2, and N and g are smooth
maps.
We will assume moreover that, for each k, ∂xk ∂ξs g αβ (ξ, x) and ∂xk ∂ξs N αβ (ξ, x) are uniformly bounded for
all |k| + s ≤ k, all r ≤ R and |ξ| ≤ 1. Here ∂xk denote multi-indices with respect to the ambient Cartesian
coordinates of Section 2.1.
For N µν (ψ, x), we will eventually need in addition to assume some version of the null condition. We will
only introduce this in Section 4.7 (see already Assumption 4.7.1). Let us note that our assumption on the
support of g − g0 is merely so as not to deal with formulations of the null condition for the quasilinear part.
We will sometimes view the equation in (4.1) as an inhomogeneous equation on a fixed g0 background,
i.e. we will write it as
g0 ψ = N µν (ψ, x)∂µ ψ ∂ν ψ + (g0 − g(ψ,x) )ψ, (4.3)
and similarly its commuted versions. We may thus apply estimates to the inhomogeneous equation
g0 ψ = F. (4.4)
4.3 Energy currents for g(ψ,x) and the stability of coercivity properties
Let ψ denote a solution of (4.1) on a domain R(τ0 , τ1 ). Because we shall use energy identities connected
with g(ψ,x) , we shall require certain basic smallness assumptions on ψ which ensure that the causal nature
of relevant hypersurfaces is retained and that induced volume forms of g and g0 are comparable.
In the sections to follow, the main smallness parameter we will consider will be εprelim > 0. We emphasise
that according to our conventions of Section 4.2, εprelim > 0 will in general depend on the nonlinearity (4.2)
(and also on k). The parameter εprelim > 0 can be taken as fixed everywhere in the paper, but note that its
smallness is constrained in multiple propositions whose preambles refer to its existence.
As we shall see, the propositions in this section will always refer to solutions satisfying
X √
e k ψ)2 ≤ ε
(D (4.5)
|k|≤1
in r ≤ R for 0 < ε ≤ εprelim . (We recall here that for r(x) ≥ R, we have g(ψ, x) = g0 .) Eventually, (4.5) will
be the consequence of a stronger estimate (see already (4.23)).
45
4.3.1 Currents for g(x,ψ) = F and the stability of coercivity properties
Let us for the moment consider more generally the equation
g(ψ,x) ψ = F (4.6)
for arbitrary F , where ψ satisfies (4.5) for sufficiently small ε.
We may define now the currents
(p) (p)
J[g(ψ,
k
x), ψ], K[g(ψ,
k
x), ψ], (4.7)
again by the expressions (3.63), where the constituent (3.10), (3.11) are defined with the same quadru-
ples (3.41) as before, but now with g = g(ψ, x) replacing g0 . These currents satisfy (3.14), with respect now
(p)
to the normals and volume forms of the metric g, and where H[ψ] k
is defined by (3.12). (Notice that the
(p)
definition of H[ψ]
k
does not depend on the metric.) In the region r ≥ R of course, all currents coincide with
their g0 versions, since g = g0 in this region.
We have the following
Proposition 4.3.1. There exists an εprelim > 0 such that the following statement holds.
Let ψ be a solution of (4.6) in R(τ0 , τ1 ) satisfying (4.5) for 0 < ε ≤ εprelim . Then x 7→ g(ψ(x), x) defines
a Lorentzian metric on R(τ0 , τ1 ) and the identity (3.14) holds in R(τ0 , τ1 , v), for all v, where the coefficients,
normals and volume forms are ε1/4 close to those of the currents (4.7) corresponding to g0 φ = F .
In particular, there exist constants C, c, such that, for p = 0 or δ ≤ p ≤ 2−δ, the corresponding coercivity
properties (3.44) are retained for the currents (4.7), while (3.42) is replaced by
(p)
K[g(ψ, x), ψ] + Ãξ(r)ψ 2 ≥ crp−1 ρ(r) (Lψ)2 + |∇ψ| / 2 + cr−1−δ ρ(r)(Lψ)2 + cr−3 ρ̃(r)ψ 2
−Cε1/4 ι{r≤R}∩{ρ≤Cε1/4 < 12 } (Lψ)2 + (Lψ)2 + |∇ψ| / 2 + ψ2 . (4.8)
Given ς = ς(k) as fixed previously, then the above applies also for the currents Jς [g(ψ, x), ψ], Kς [g(ψ, x), ψ]
in the region r ≥ rKilling , while in the region r ≤ r1 (ς), the coercivity properties of Propositions 3.4.1 and 3.4.2
hold for these currents.
Proof. This is clear from the properties of tensor identities. Note how the smallness constraint on εprelim
provided by this proposition depends of course on the map g of (4.2) and is needed even simply to ensure
for g to be Lorentzian and for the inverse metric to be well defined.
Note that since ρ = ρ̃ = 1 in case (i), the bound (4.8) implies that the full coercivity applies in that
case. (Note that if ρ is a step function valued in {0, 1}, then {ρ ≥ Cε1/4 } = {ρ ≥ 12 } = {ρ = 1} and
{ρ ≤ Cε1/4 } = {ρ = 0}.)
Corollary 4.3.2. Under the assumption of Proposition 4.3.1, we have the coercivity statement
(p)
8 (p)
|k|≤k−1
46
Proof. Note that we have retained g0 on the left hand side of (4.9), so this is simply a statement about the
stability of coercivity properties of the first term on the right hand side.
In view of the above we will define a new set of energy quantities, F(g), E(g),
k
etc., defined in parallel with
k
(0) (p) (0) (p)
those of F, Ek
of Section 3.6.7 (to be denoted below as F(g0 ), E(g k
0 )), but where the flux is defined with
k k
respect to the divergence identity with respect to g = g(ψ, x), i.e. we define
Z(p) (p)
Z (p) (p)
µ
E(g,
k
τ ) := J
k µ
[g, ψ]n(g) Σ(τ ) , Ek S
(g, τ ) := Jk µ [g, ψ]n(g)µS ,
Σ(τ ) S
Z(p)
(4.11)
(p)
where the omitted volume forms above are here understood with respect to g = g(ψ, x).
Corollary 4.3.3. We have that under the assumptions of Proposition 4.3.1,
(p) (p) (p) (0) (p) (p) (0) (0)
E(g,
k
τ ) ∼ E(g
k
0 , τ ) . E(τ
k
), ES (g, τ ) = ES (g, τ ) ∼ E
k S
(g0 , τ ) = E
k S
(g0 , τ ) ∼ Ek S (τ ),
(p) (p) (p)
F(g, v) = F(g0 , v) ∼ F,
k
k k
in fact
(0) (0) (0) (0)
1 1
E(g)
k
≤ (1 + Cε 4 )E(g
k
0 ), E(g
k
0 ) ≤ (1 + Cε 4 )E(g),
k
(p)
Z
(p) X
E(τ ) . E[g](τ )+ e k F + (g − g(ψ,x) ))ψ)2 .
(D (4.12)
k k 0
Σ(τ )∩{r∗ ≤r≤r 00 } |k|≤k−1
(p)
As always, the significance of expressing the above with respect to the energies E(g),
k
etc., is that the
constants in our nonlinear inequalities will be exactly 1.
Note that in all integrals below with volume form omitted, we will continue to use the volume form
induced by g0 unless otherwise noted. (We shall only use the volume form induced by g(ψ, x) when applying
the divergence identity involving (4.11). The corresponding volume forms are of course equivalent under
the assumptions of Proposition 4.3.1, but again one must distinguish where appropriate so as to obtain the
exact constant.)
Proposition 4.4.1. For all k ≥ 1. There exists an εprelim > 0 such that the following statement holds.
Let 0 < δ ≤ p ≤ 2 − δ, τ0 ≤ τ1 and let ψ be a solution of (4.1) in R(τ0 , τ1 ) satisfying (4.5) for
47
0 < ε ≤ εprelim . Then for all τ ∈ [τ0 , τ1 ], we have
(p) (p)
Z τ (p)
Z (p−1)
τ (−3−δ)
ρ 0 0
sup F(v, τ0 , τ ), E(τ
k
) + E
k S
(τ 0 , τ ) + c E
k
(τ )dτ 0
+ c ρ̃
k−1
E 0 (τ 0 )dτ 0
k
v:τ ≤τ (v) τ0 τ0
(p)
Z τ ξ
≤ E(τ
k
0 ) + A E 0 (τ 0 )dτ 0
k−1
τ0
Z (p)
+ H[ψ]
k
· {Dk (N αβ (ψ, x)∂α ψ∂β ψ)} (4.13)
R(τ ,τ )
Z 0 (p)
+ H[ψ]
k
· {[g(ψ,x) − g0 , Dk ]ψ} (4.14)
R(τ0 ,τ )∩{r≤R}
Z Z
X X 2
+C (Dk (N αβ (ψ, x)∂α ψ∂β ψ))2 + C [g(ψ,x) − g0 , Dk ]ψ
R(τ0 ,τ ) |k|≤k R(τ0 ,τ )∩{r≤R} |k|≤k
Z X
+C e k (N αβ (ψ, x)∂α ψ ∂β ψ + (g − g(ψ,x) )ψ))2
(D (4.15)
0
R(τ0 ,τ )∩{r≤R} |k|≤k−1
Z X
1/4 e k ψ))2 + |∇(
e k ψ))2 + (L(D / De k ψ)|2 + ψ 2 (4.16)
+C ε √
(L(D
R(τ0 ,τ )∩{r≤R}∩{ρ≤C ε} |k|≤k
(4.18)
Z X 2
+ Dk (N αβ (ψ, x)∂α ψ ∂β ψ + (g0 − g(ψ,x) )ψ) (4.19)
R(τ0 ,τ ) |k|≤k−1
Z X 2
+ e k (N αβ (ψ, x)∂α ψ ∂β ψ + (g − g(ψ,x) )ψ)
D (4.20)
0
R(τ0 ,τ )∩{r≤R} |k|≤k−2
Z X 2
+ e k (N αβ (ψ, x)∂α ψ ∂β ψ + (g − g(ψ,x) )ψ)
D . (4.21)
0
Σ(τ )∩{r≤R} |k|≤k−2
+ H[ψ]
k
· {[g0 , Dk ]ψ}
R(τ0 ,τ )∩{r≤R}
and then bringing the second term to the left hand side to use Corollary 4.3.2 and absorb it in the bulk.
(Recall that in the region r ≥ R, we have g = g0 .) The term on line (4.15) arose from our application of
elliptic estimates to ψ, considering it as a solution of (4.3). The term on line (4.16) arose from the term on
the last line of (4.8). We note that restricted to r ≥ R, we may replace De commutation with D commutation.
The inequality (4.17)–(4.21), on the other hand, simply arises from applying the estimate (3.109) to the
nonlinear equation written in the form (4.3). Note that we have applied it at one order less, i.e. with k − 1
in place of k, in view of the fact that the right hand side is of order k.
48
4.5 Summed norm notation and the lower order smallness assumption
In addition to the primitive assumption (4.5), in the context of our proof, we will need to introduce stronger
a priori energy smallness assumptions on ψ in our region R(τ0 , τ1 ) under consideration. In the context of the
proof of the main theorem, this will appear as a bootstrap assumption. This will ensure that higher order
nonlinear terms are indeed absorbable.
We first explain some additional notation. We define the following summed quantities for δ ≤ p ≤ 2 − δ:
(p) (p)
Z τ1 (p) (p−1)
X
k
(τ ,
0 1τ ) := sup E(τ
k
0
) + sup F(v,
k
τ ,
0 1τ ) + Ek 0 (τ 0 )dτ 0 ,
τ 0 ∈[τ0 ,τ1 ] v:τ1 ≤v(τ ) τ0
(p) (p) (p)
Zτ1 (p−1) (−3−δ)
ρ 0
ρ
X
k
(τ0 , τ1 ) := sup E(τ
k
0
)+ sup F(v,
k
τ0 , τ1 ) + E 0 (τ 0 ) dτ 0 ,
Ek (τ 0 ) + ρ̃k−1
τ 0 ∈[τ 0 ,τ1 ] v:τ1 ≤v(τ ) τ0
(p) (p) (p)
Zτ1 (p−1) (p−1)
χ 0
χ
X
k
(τ0 , τ1 ) := sup E(τ
k
0
)+ sup F(v,
k
τ0 , τ1 ) + E 0 (τ 0 ) dτ 0 .
Ek (τ 0 ) +k−1
τ 0 ∈[τ0 ,τ1 ] v:τ1 ≤v(τ ) τ0
X
k
(τ ,
0 1τ ) := sup E(τ
k
0
) + sup F(v,
k
τ ,
0 1τ ) + Ek 0 (τ 0 )dτ 0 ,
τ 0 ∈[τ0 ,τ1 ] v:τ1 ≤v(τ ) τ0
(0) (0) (0)
Zτ1 (−1−δ) (−3−δ)
ρ 0
ρ
X
k
(τ0 , τ1 ) := sup E(τ
k
0
)+ sup F(v,
k
τ0 , τ1 ) + E 0 (τ 0 ) dτ 0 ,
Ek (τ 0 ) + ρ̃k−1
τ 0 ∈[τ0 ,τ1 ] v:τ1 ≤v(τ ) τ0
(0) (0) (0)
Zτ1 (−1−δ) (−1−δ)
χ 0
χ
X
k
(τ0 , τ1 ) := sup E(τ
k
0
)+ sup F(v,
k
τ0 , τ1 ) + E 0 (τ 0 ) dτ 0 .
Ek (τ 0 ) +k−1
τ 0 ∈[τ0 ,τ1 ] v:τ1 ≤v(τ ) τ0
Because p = 0 is anomalous, however, we will need in addition the following stronger energies which will
appear on the right hand side of p = 0 estimates:
(0+) (0)
Z τ1 (0) (δ − 1)
Xk
(τ ,
0 1τ ) := sup E(τ
k
0
) + sup F(v,
k
τ ,
0 1τ ) + Ek 0 (τ 0 )dτ 0 ,
τ 0 ∈[τ0 ,τ1 ] v:τ1 ≤v(τ ) τ0
(0+) (0) (0)
Z τ1 (δ − 1) (−3−δ)
ρ 0
ρ
X
k
(τ0 , τ1 ) := sup E(τ
k
0
)+ sup F(v,
k
τ0 , τ1 ) + E 0 (τ 0 ) dτ 0 ,
Ek (τ 0 ) + ρ̃k−1
τ 0 ∈[τ0 ,τ1 ] v:τ1 ≤v(τ ) τ0
(0+) (0) (0)
Z τ1 (δ − 1) (δ − 1)
χ 0
χ
X
k
(τ0 , τ1 ) := sup E(τ
k
0
)+ sup F(v,
k
τ0 , τ1 ) + E 0 (τ 0 ) dτ 0 .
Ek (τ 0 ) +k−1
τ 0 ∈[τ0 ,τ1 ] v:τ1 ≤v(τ ) τ0
X . χX
k−1 k
. (4.22)
We will use the notation k to denote some particular positive integer, depending on k, which may be
different on different instances of our use of the notation, such that k ≤ k, and in fact, k is “much less
than k”, provided k is sufficiently large. In particular, for all positive integers n, we assume there exists a
k(n) for which k ≥ k(n) implies k ≤ k − n.
We have the following:
Proposition 4.5.1. Let k ≥ 4 be sufficiently large. There exists an εprelim > 0 such that for all 0 < ε ≤
εprelim , the following holds.
Let ψ satisfy
(p)
X ≤ε
k
(4.23)
in R(τ0 , τ1 ), for p = 0 (or some δ ≤ p ≤ 2 − δ), where k ≥ 4. Then the following improved version of (4.5)
holds: X √
sup (De k ψ)2 . ε ≤ ε (4.24)
r≤R
|k|≤1
49
Proof. This follows of course immediately from the Sobolev inequality (3.111).
In the context of the proof of the main theorem, inequality (4.23) will be introduced as a bootstrap
assumption, and Proposition 4.5.1 will be applied to retrieve the assumption (4.5), necessary for the results
√
of Sections 4.3–4.4. Note that with the assumption (4.23), we may replace the boxed ε1/4 in (4.16) with ε.
We note that in what follows, restrictions on k sufficiently large will always include the condition k ≥ 4.
Proof. This is standard in view of our assumptions on N and g(ψ, x) from Section 4.1, and can be proven
using the Sobolev inequality (3.111) on Σ(τ 0 ) ∩ {r ≤ R}.
We may now obtain the following result, which can be viewed as a corollary of Proposition 3.6.6.
Corollary 4.5.3. Let k ≥ 4 be sufficiently large. Then there exists an εprelim > 0 such that the following
statement holds.
Let ψ be a solution of (4.1) in R(τ0 , τ1 ) satisfying (4.23) for p = 0 and some 0 < ε ≤ εprelim . We have
the analogue of Corollary 3.6.7: The calligraphic and fraktur (when applicable) energies are comparable, i.e.
(p) (p)
E(τ
k
) ∼ E(τ
k
)(g),
for all τ0 ≤ τ ≤ τ1 . Moreover, if χ = 1 and ρ̃ = 1 identically (as in case (i)), then
(p−1) (p−1) (−3−δ) (−1−δ) (−1−δ) (−3−δ)
Ek 0 (τ ) ∼ χ Ek 0 (τ ) + ρ̃k−1
E 0 (τ ), Ek 0 (τ ) ∼ χ Ek 0 (τ ) + ρ̃k−1
E 0 (τ )
and thus (p) (p) (p)
χ
X
k
(τ0 , τ1 ) ∼ ρ X
k
(τ0 , τ1 ) ∼ X
k
(τ0 , τ1 ).
If ρ̃ = 1 identically, ρ = χ, (i.e. as in cases (i) and (ii)), then
(p−1) (p−1) (p−1) (−3−δ) (−1−δ) (−1−δ) (−1−δ) (−3−δ)
χ 0
E 0 (τ ) ∼ ρ Ek 0 (τ ) + ρ̃k−1
Ek (τ ) +k−1 E 0 (τ ), χ 0
E 0 (τ ) ∼ ρ Ek 0 (τ ) + ρ̃k−1
Ek (τ ) +k−1 E 0 (τ )
and thus (p) (p)
χ
X
k
(τ0 , τ1 ) ∼ ρ X
k
(τ0 , τ1 ).
Proof. This follows from (3.103)–(3.107), with F defined as in Proposition 4.5.2, absorbing the inhomoge-
neous term using the estimate (4.25), for sufficiently small εprelim .
... . E(τ
k
0
) E 0 (τ 0 ) dτ 0 ,
k
(4.26)
τ0
while the integrals on lines (4.18)–(4.19), again restricted to R(τ0 , τ1 ) ∩ {r ≤ R}, may be similarly estimated
by
Z τ q (0) (−1−δ)
... . E(τ
k
0
) E 0 (τ 0 ) dτ 0 .
k
(4.27)
τ0
50
Proof. This is standard given the assumptions in Section 4.1 and can be proven using the Sobolev inequal-
ity (3.111) on Σ(τ ) ∩ {r ≤ R}, cf. the proof of (4.25) which in fact already bounds
q several of the terms. (Note
(−1−δ)
that the boxed ε1/4 in (4.16) may clearly now be replaced by the quantity E 0 (τ 0 ) within the integral.)
k
We remark again that, as opposed to (4.26), the control (4.27) loses a derivative, as the energy on the
left hand side of (4.18) is k − 1’th order while the right hand side of (4.27) is of k’th order.
4.7 Assumptions for the nonlinearity in the far region: the “null” condition
We may now state our specific assumption capturing the null condition for the semilinear terms.
Rather than formulate algebraically the null condition in terms of the form of N , for maximum generality
we will make our basic assumption directly at the level of an estimate for terms on the right hand side of
the inequalities of Proposition 4.4.1.
Our assumptions will only depend on the region r ≥ 8R/9, so we will need some notation to denote the
restriction of energies, etc., to this region. Given r∗ ≥ r0 and v, by
(p) (p) (p)
ρ
Ek r , X , X , ...
∗ ,v k r ∗ ,v k r ∗ ,v
we shall mean the expressions defined as before, but where all domains of integration are restricted to the
region R(τ0 , τ1 , v) ∩ {r ≥ r∗ }. We may thus define these expressions for functions ψ : R(τ0 , τ1 , v) ∩ {r ≥
(p)
r∗ } → R. Similarly, the above expressions without the v subscript, e.g. Ek r , will be defined where all
∗
domains of intergation are restricted to R(τ0 , τ1 ) ∩ {r ≥ r∗ }. These can then be defined for functions
ψ : R(τ0 , τ1 ) ∩ {r ≥ r∗ } → R . Let us note that
(p) (p) (p)
r∗ ≥ 8R/9 =⇒ X
k r
= χX = ρX (4.28)
∗ ,v k r ∗ ,v k r ∗ ,v
k
X 8R/9,v ≤ ε (4.29)
.X
k
8R X 8R ,v (τ0 , τ1 ) + X
(τ , τ ) k
,v 0 1 k
8R (τ , τ ) X
,v 0 1 k
8R
,v 0 1
X 8R ,v (τ0 , τ1 ), (4.30)
(τ , τ ) k
9 9 9 9 9
.X
k
8R
,v 0 1
X 8R ,v (τ0 , τ1 ).
(τ , τ ) k (4.31)
9 9
51
Proof. See Appendix C.
Remark 4.7.3. Let us note that for ψ defined on the slab R(τ1 , τ2 ) ∩ {r ≥ 8R/9}, from the trivial bound
(p) (p)
X
k r ,v
≤X
k r
we may drop the v subscripts. For ψ defined globally on the slab R(τ1 , τ2 ), then in view of (4.28)
∗ ∗
(p) (p) (p)
8M
and the trivial bounds ρ X
k r
≤ ρX
k
, etc., we may replace the 9 ,v subscripts in all the X
k
, etc., energies on
∗
(p) (p)
sup F(v, τ0 , τ ), c ρX
k
(τ0 , τ1 ), E(τ
k
1)
v:τ ≤τ (v)
(p)
Z τ1 ξ
≤ E(τ
k
0) + A E 0 (τ 0 )dτ 0
k−1
τ0
r r r r
(p) (0) (p) (0) (p)
+CX k
8R (τ0 , τ1 ) X (τ0 , τ1 ) + C X
k 8R k
8R (τ0 , τ1 ) X
k
8R (τ0 , τ1 ) X (τ0 , τ1 )
k 8R
9 9 9 9 9
Z τ1 q (0) (−1−δ)
+C E(τ
k
0
) k E 0 (τ 0 ) dτ 0 (4.32)
τ0
r r r r
(p) (p) (p) (0) (p) (0) (p)
χ
X
k−1
(τ ,
0 1τ ) . E(τ
k−1
0 ) + X
k−1 8R
(τ ,
0 1 τ ) X
k 8R
(τ ,
0 1τ ) + X
k−1 8R
(τ ,
0 1τ ) X
k−1 8R
(τ ,
0 1τ ) X (τ0 , τ1 )
k 8R
9 9 9 9 9
Z τ1 q (0) (−1−δ)
+ E(τ
k
0
) k E 0 (τ 0 ) dτ 0 (4.33)
τ0
Finally, in both the case p = 0 and the case δ ≤ p ≤ 2 − δ, one has the alternative estimate
(p)
q (p) (p) (0)
X
k
(τ0 , τ1 ) . E(τ
k
0 ) + (1 + τ1 − τ0 )X
k
X (τ0 , τ1 ),
(τ0 , τ1 ) k (4.36)
depending on τ1 − τ0 .
Proof. This follows from Proposition 4.4.1, Corollary 4.5.3, Proposition 4.6.1 and Assumption 4.7.1.
Remark 4.8.2. Note how the case p = 0 is anomalous in that one sees 0+ on the right hand side of (4.34)–
(4.35). It is this fact which will require us to use p > 0 weights to close the global estimates (4.32)–(4.33),
even in the case (i). On the other hand, (4.36) is sufficient to show local existence even using only the p = 0
weight.
52
(p) (p)
ρ
Remark 4.8.3. In using the above we shall always simply replace terms like X
k
8R (τ0 , τ1 ), etc., by X
k
(τ0 , τ1 ),
9
etc., according to Remark 4.7.3. We have kept the terms in the above form simply to highlight their origin
in the estimate of Assumption 4.7.1.
Remark 4.8.4. In the case where we only assume the black box inequality (3.3) and the assumptions of
asymptotic flatness of Sections 3.5 and 3.6.2 as interpreted in Remark 3.6.9 (necessary to obtain (3.109))
and moreover, the equation (4.1) is semilinear, i.e. g(x, ψ) = g0 for all x ∈ M, then we obtain the statement
of Proposition 4.8.1 without inequalities (4.32) and (4.34) and where (4.33) and (4.35) are replaced by
r r r r
(p) (p) (p) (0) (p) (0) (p)
χ
X
k
(τ0 , τ1 ) . E(τ
k
0) + X k
8R (τ0 , τ1 ) X (τ0 , τ1 ) + X
k 8R k
8R (τ0 , τ1 ) X
k
8R (τ0 , τ1 ) X (τ0 , τ1 )
k 8R
9 9 9 9 9
Z τ1 (0)
q (−1−δ)
+ E(τ
k
0
) k E 0 (τ 0 ) dτ 0 , (4.37)
τ0
(0) (0) (0+)
r
(0+)
Z τ1 q (0) (0)
0
χ
X
k
(τ0 , τ1 ) . E(τ
k
0 ) + X
k
8R (τ 0 , τ1 ) X
k 8R
(τ 0 , τ1 ) + E(τ
k
) E(τ 0 ) dτ 0 ,
k
(4.38)
9 9 τ0
E[ψ,
k
ψ0 ] (4.39)
as follows:
Using the equation (4.1) we may compute along Σ(τ0 ) the k + 1-jet of any smooth solution ψ of (4.1)
such that
ψ|Σ(τ0 ) = ψ, nψ|Σ(τ0 )∩{r≤R} = ψ0 . (4.40)
(p) (p)
define (4.39) substituting this Ψ for ψ in the usual expression for E(τ k
0 ).
With this, we can now state our local well posedness:
Proposition 4.9.1 (Local well posedness). Consider (M, g0 ) satisfying the assumptions of Sections 2 and 3
(for cases (i), (ii) or (iii)) and equation (4.1) satisfying the assumptions of Section 4.1 and 4.7. Fix either
p = 0 or δ ≤ p ≤ 2 − δ.
There exists a positive integer kloc ≥ 4 such that the following holds. Let k ≥ kloc . Then there exists
a positive real constant C > 0 sufficiently large, a positive real parameter εloc > 0 sufficiently small, and a
decreasing positive function τexist : (0, εloc ) → R such that for all smooth initial data (ψ, ψ0 ) on Σ(τ0 ) such
that (p)
E[ψ,
k
ψ0 ] ≤ ε0 ≤ εloc ,
there exists a smooth solution ψ of (4.1) in R(τ0 , τ1 ) for τ1 := τ0 + τexist which satisfies
(p)
X
k
(τ0 , τ1 ) ≤ Cε0 .
53
Moreover, for all τ0 ≤ τ ≤ τ1 , any other smooth ψe defined on R(τ0 , τ ) satisfying (4.1) in R(τ0 , τ ) and
attaining the initial data, i.e. satisfying (4.40) (with ψe replacing ψ), coincides with the restriction of ψ,
i.e. ψe = ψ|R(τ0 ,τ ) .
We have in addition the following propagation of higher order regularity and/or higher weighted estimates.
Given p, k ≥ kloc , ε0 ≤ εloc , τ0 , τ1 as above, 2 − δ ≥ p0 ≥ p if p ≥ δ and either 2 − δ ≥ p0 ≥ δ or p0 = 0 if
p = 0, and k 0 ≥ k. Then there exists a constant C(k 0 , τ1 − τ0 ) such that given ψ as above,
(p0 ) (p0 )
X
k0 0
(τ0 , τ1 ) ≤ C(k 0 , τ1 − τ0 ) kE0 0 (τ0 ).
r∗ r∗
Proof. This can be easily proven using estimate (4.36). We leave the details to the reader.
Note also the following easy corollary, which we will use as an extension criterion:
Corollary 4.9.2 (Continuation criterion). Fix p = 0 or δ ≤ p ≤ 2 − δ, and let k ≥ kloc and εloc be as in
Proposition 4.9.1. There exists a constant C > 0 and an > 0 such that the following is true.
Let τf > τ̃0 and suppose that there exists a smooth solution ψ of (4.1) on ∪τ̃0 <τ <τf R(τ̃0 , τ ) such that
(p)
E(τ
k
) ≤ εloc
for k ≥ kloc and all τ̃0 ≤ τ < τf . Then, defining τ1 := τf + , ψ extends uniquely to a smooth function on
R(τ̃0 , τ1 ) satisfying the equation (4.1) on this set and, setting τ0 := max{τf − , τ̃0 }, satisfying the estimate
(p)
X
k
(τ0 , τ1 ) ≤ Cεloc . (4.41)
Moreover, for all k 0 ≥ k and all 2 − δ ≥ p0 ≥ p, there exist constants C(k 0 ) such that
(p0 ) (p0 )
X
k0
(τ0 , τ1 ) ≤ C(k 0 ) kE(τ
0 0 ). (4.42)
Remark 4.9.3. Recall that by our conventions the constants C, kloc , εloc depend in addition on k. We
may assume that kloc is sufficiently large so that k ≥ kloc satisfies the largeness constraint of all previous
propositions in this Section, and that εloc ≤ εprelim .
Remark 4.9.4. In view of Remark 4.8.4, and the fact that one needs only (4.36), all the statements of this
section also hold in the semilinear case, under the relaxed assumptions described there.
E[ψ,
k
ψ0 ] ≤ ε0 ≤ εglobal , E[ψ,
k
ψ0 ] +k−1
E[ψ, ψ0 ] ≤ ε0 ≤ εglobal , E[ψ,
k
ψ0 ] +k−2
E[ψ, ψ0 ] ≤ ε0 ≤ εglobal ,
according to case (i), (ii) and (iii) respectively, where in the former case δ ≤ p ≤ 2 − δ. Then the ψ given by
Proposition 4.9.1 extends to a unique globally defined solution of (4.1) in R(τ0 , ∞), satisfying the estimates
(p) (1) (2 − δ) (1) (2 − δ)
ρ ρ
X
k
(τ0 , τ ) ≤ Cε0 , X
k
(τ0 , τ ) + ρk−1
X (τ0 , τ ) ≤ Cε0 , ρ
X
k
(τ0 , τ ) + ρk−2
X (τ0 , τ ) ≤ Cε0 , (5.1)
according to cases (i), (ii), (iii), respectively, for all τ ≥ τ0 , in particular
(p) (1) (2 − δ) (1) (2 − δ)
E(τ
k
) ≤ Cε0 , E(τ
k
E(τ ) ≤ Cε0 ,
) +k−1 E(τ
k
E(τ ) ≤ Cε0 ,
) +k−2
respectively.
The solution will satisfy moreover estimates (6.8) in case (i), estimates (6.62) and (6.64)–(6.68) in case
(ii) and estimates (6.116)–(6.121) in case (iii).
54
We shall give the proof of this theorem in Section 6.
Remark 5.1. Our theorem also applies under the relaxed assumptions of Remark 4.8.4, where (4.1) is
however required to be semilinear. Here we can again distinguish case (i) where χ = 1 and case (ii) where χ
is allowed to degenerate. In case (i), the theorem holds as stated except for (6.8), which no longer applies.
Similarly, in case (ii), the theorem holds as stated except for (6.62) which no longer applies. The necessary
modifications will be collected in a series of remarks in Sections 6.1 and 6.2. (See already Section 6.2.5 for
the completion of case (ii).)
We reiterate (cf. Propositions 4.7.2 and Theorem A.1) that the above theorem in particular holds for
equations (4.1) on the Kerr spacetime (M, ga,M ) for |a| M , with quasilinear terms as described in
Section 4.1 and the general semilinear terms considered in [Luk13], and, by the above remark, restricted to
the purely semilinear case, for the full range of parameters |a| < M .
Proof. Let εboot ≤ εloc . The assumption of Proposition 4.8.1 is satisfied in view of our discussion follow-
ing (6.1). From estimate (4.32) of Proposition 4.8.1, since in case (i) we have ρ = 1, A = 0, Ã = 0, it follows
that q q
(p) (p) (p) (p) (p) (p) (p) (p)
X
k
. E(τ
k
0 ) + X
k
X
k
, sup E(τ
k
) ≤ E(τ
k
0 ) + C X
k
X.
k
(6.3)
τ0 ≤τ ≤τ1
55
By our bootstrap assumption (4.23), we have
(p)
X ≤ ε.
k
(6.4)
Thus, possibly requiring εboot to be even smaller, we may absorb the nonlinear term in the first inequality
of (6.3) to obtain the first inequality of (6.2).
To obtain the second inequality of (6.2), we remark that Corollary 4.5.3 applies to yield in particular
(p) (p)
E(τ
k
0 ) ∼ E(τ
k
0 ). (6.5)
The second inequality of (6.2) now immediately follows from the the second inequality of (6.3), plugging in
the first inequality of (6.2) just established, (6.5) and (6.4).
Remark 6.1.2. Under the relaxed assumptions of Remark 4.8.4, where (4.1) is however required to be
semilinear and we are in the analogue of case (i), i.e. χ = 1, then from (4.38) we obtain the first inequality
of (6.3). Thus, the above proposition holds as stated for the first inequality of (6.2).
E(τ
k
0 ) ≤ ε0 ≤ εglobal
X
k
. E(τ
k
0 ) . ε0 . (6.7)
If follows that for sufficiently small εglob , and all ε0 ≤ εglob , the continuation criterion Corollary 4.9.2
applies to obtain that ψ extends as a solution to (4.1) on some R(τ0 , τf +), where is independent of τf , and
that, considering now τ1 := τf + , inequality (6.7) holds also in R(τ0 , τ1 := τf + ) (with a slightly different
(p)
implicit constant than that of (6.7)). (To infer this from (4.41), which was an estimate on X k
(τf , τf + ),
(p) (p) (p)
note that Xk
(τ0 , τ1 ) . X
k
(τ0 , τf ) + X
k
(τf , τf + ).)
Finally we note that for ε satisfying (6.6), inequality (6.7) in R(τ0 , τ1 := τf + ) shows in particular that
inequalities (4.23) hold in this set. It follows by the definition of B that τf + ∈ B and thus, given also its
connectedness, the set B is open. Since does not depend on τf , it follows that B is also closed, as any limit
point τlimit of B in (τ0 , ∞) satisfies τlimit ≤ τf + for some τf ∈ B. We have shown that B is a non-empty
open and closed subset of (τ0 , ∞) and thus, B = (τ0 , ∞). Hence the solution ψ exists globally in R(τ0 , ∞)
and satisfies (6.7) in R(τ0 , τ1 ) where τ1 is now any τ1 > τ0 . This gives (5.1).
Revisiting (6.2) we see finally that we have the following more precise global estimate:
(p) (p)
1/2
E(τ
k
) ≤ E(τ
k
0 )(1 + Cε0 ). (6.8)
This completes the proof.
Remark 6.1.3. Note that in view of Remark 6.1.2, the above proof goes through, except for the last paragraph
involving (6.8), also under the relaxed assumptions of Remark 4.8.4, where (4.1) is however required to be
semilinear and we are in the analogue of case (i), i.e. χ = 1. Thus, in this case, one indeed obtains the
statement as given in Remark 5.1.
56
6.2 Case (ii)
Case (ii) is the second simplified case which we shall consider. Unlike case (i), we shall require proving actual
τ -decay above a certain threshold in order to close, and this in turn will require raising p to p = 2 − δ in our
initial energy assumptions.
The logic of the proof is a simplified version of the scheme we shall use for the general case and which
has already been summarised in the introduction:
1. Rather than bootstrap directly t-weighted estimates, we formulate the fundamental estimates, depend-
ing only on the bootstrap assumption (4.23), as a hierarchy of estimates on a spacetime slab of τ -length
at most L. This is the content of Section 6.2.1.
2. We shall then show by a bootstrap argument, restricted to such a slab, how global existence and
estimates on the slab can be proven, with suitable assumptions on the initial data of the slab, which
now involve L. This is the content of Section 6.2.2.
3. Still restricted to a given slab of length L, we will show that one can in fact a posteriori improve the
above estimates under suitable additional assumptions on the initial data, and, using a pigeonhole
argument, show moreover an improved estimate for any Σ(τ 0 ) slice near the top of the slab. This is
the content of Section 6.2.3.
4. Finally, global existence and τ decay now follow by iterating the above estimates on a consecutive
sequence of spacetime slabs of dyadic time-length Li = 2i . This is the content of Section 6.2.4.
We shall in addition give an alternative iteration proof in Section 6.2.5, which does not use the exact
boundedness statement of the fraktur energies. The advantage of this alternative proof is that it allows us
to treat the semilinear case with the relaxed assumptions of Remark 4.8.4. The modifications necessary to
treat this case are described in a series of remarks (see already Remarks 6.2.2, 6.2.4, 6.2.7 and 6.2.8).
τ1 ≤ τ0 + L
for some arbitrary L > 0. We have the following hierarchy of inequalities on R(τ0 , τ1 ):
(2 − δ) (2 − δ) (2 − δ)
q
(2 − δ)
q q q (2 − δ)
q √ (0) (2 − δ) (0) (2−δ) (0) (0)
χ χ χX χX χ
F(v, τ1 ), E(τ
k
1 ), c X
k
≤ E(τ
k
0 ) + C X
k
X
k
+ k k
X
k
+ C X
k
X L,
k
(6.9)
k
+C χ Xk
X L.
k
(6.11)
Proof. Again we recall that if εboot ≤ εloc , then the assumption of Proposition 4.8.1 holds. The first two
inequalities follow from the estimate (4.32) of Proposition 4.8.1 applied to p = 2 − δ, p = 1 in view of the fact
that ρ = χ, where we have used the relations (4.22) and (4.28) to replace the far-away supported nonlinear
terms with those displayed above, together with the estimate
sZ
Z τ1
0
q
(0) (−1−δ)
0 0 0
τ1
0 0
(0) √ χ
q √ (−1−δ) (0) (0)
57
q (1) (1)
Note that for the p = 1 estimate we retain only the less precise expression χ X k
X which will be sufficient
k
Following the above proof but now using equations (4.37) and (4.38), we may in fact directly deduce the
inequalities (6.12)–(6.14) under the relaxed assumptions of Remark 4.8.4, where (4.1) is however required to
be semilinear and we are in the analogue of case (ii). Thus, the analogue of Proposition 6.2.1 holds in that
case where (6.9)–(6.11) are replaced by (6.12)–(6.14).
E(τ
k
0 ) ≤ ε0 , E(τ0 ) ≤ ε0 L−1 ,
k−1
(6.15)
then the unique solution of Proposition 4.9.1 obtaining the data can be extended to a ψ defined on the entire
spacetime slab R(τ0 , τ0 + L) satisfying the equation (4.1) and the estimates
(1) (0)
χ
X
k
. ε0 , χ
X . ε0 L−1 .
k−1
(6.16)
Proof. Consider the set B ⊂ (τ0 , τ0 + L] consisting of all τ0 + L ≥ τf ≥ τ0 such that a solution ψ of (4.1)
obtaining the data exists on R(τ0 , τf ) and such that the boostrap assumption (4.23) with p = 1 and also the
additional bootstrap assumption
(0)
X ≤ εL−1
k
(6.17)
hold in R(τ0 , τ1 := τf ), where 0 < ε ≤ εboot is a small constant satisfying
1 ε εslab . (6.18)
58
where we have used the bootstrap assumptions (4.23) and (6.17).
From (6.11) for k − 1 and (6.15), we obtain
(0) (0) 1−δ (1)
2δ 1 1−δ (0)
χ
k−1
X . ε0 L−1 + ε1/2 ( χk−1 X ) 1+δ L− 2 1+δ + ε1/2 χk−1
X ) 1+δ ( χk−1 X. (6.20)
Now defining
2
1 1−δ 11−δ
γ0 := + ,
2 1+δ 21+δ
we have (0)
χ
k−1
X . ε0 L−γ0 , (6.23)
whence, plugging (6.23) into (6.20) we improve to
(0) 1−δ 2 1 1−δ
χ
k−1
X . ε0 L−1 + ε1/2 ε0 L−γ0 ( 1+δ ) L− 2 1+δ . (6.24)
2
Defining γi iteratively by γi = γi−1 1−δ
1+δ + 1 1−δ
2 1+δ , then, in view of the restriction (3.1), there exists a first
i such that γi ≥ 1, whence we obtain
(0)
χ
X . ε0 L−1 .
k−1
(6.25)
We may now already apply our continuation criterion Corollary 4.9.2, applied with p = 0 and k − 1, to
assert the existence of an , independent of τ1 , such that now defining τ1 := min{τf + , τ0 + L}, the solution
ψ extends to a smooth solution of (4.1) on R(τ0 , τ1 ), and moreover, from (4.41), that
(0)
χ
X . ε0 L−1
k−1
(6.26)
for a sufficiently small β, in which case the first inequality of (6.16) is replaced by
(1)
χ
X
k
(τ0 ) ≤ ε0 Lβ . (6.29)
59
6.2.3 The pigeonhole argument
The above assumptions on initial data are sufficient for global existence in the slab, but are not sufficient to
iterate. For this we shall need strengthened assumptions.
Proposition 6.2.5. Under the assumptions of Proposition 6.2.3, there exists a constant C > 0, implicit in
the inequalities . below, a parameter α0 1 and, for all α ≥ α0 , a parameter ε̂slab (α) such that for all
0 < ε̂0 ≤ ε̂slab (α) the following holds.
Let us assume that in addition to (6.15) we have
(1) (2 − δ) (0) (1) (0)
E(τ
k
0 ) ≤ ε̂0 , E(τ0 ) ≤ ε̂0 ,
k−1
E(τ0 ) ≤ ε̂0 αL−1 ,
k−1
E(τ0 ) ≤ ε̂0 αL−1+δ ,
k−2 k−3
E(τ0 ) ≤ ε̂0 α2 L−2+δ . (6.30)
Then the solution ψ of (4.1) on R(τ0 , τ0 + L) given by Proposition 6.2.3 satisfies the additional estimates
(1)
sup E(τ
k
) ≤ αε̂0 , (6.31)
τ0 ≤τ ≤τ0 +L
(1)
1/4
sup E(τ
k
) ≤ ε̂0 (1 + ε̂0 L(−1+δ)/2 ), (6.32)
τ0 ≤τ ≤τ0 +L
(2 − δ)
1/4
sup E(τ ) ≤ ε̂0 (1 + ε̂0 L−1/2 ),
k−1
(6.33)
τ0 ≤τ ≤τ0 +L
(1) (2 − δ) (0) (1) (0)
χ
X
k
. ε̂0 , χ
k−1
X . ε̂0 , χ
X . ε̂0 αL−1 ,
k−1
χ
k−2
X . ε̂0 αL−1+δ , χ
k−3
X . ε̂0 α2 L−2+δ . (6.34)
Moreover, for all times τ 0 with L ≥ τ 0 − τ0 ≥ L/2, we have that
(0)
1
E(τ 0 ) ≤ ε̂0 αL−1 , (6.35)
k−1
2
(1)
1
E(τ 0 ) ≤ ε̂0 αL−1+δ , (6.36)
k−1
2
(0)
1
E(τ 0 ) ≤ ε̂0 α2 L−2+δ . (6.37)
k−2
4
Proof. Note that by the statement of the proposition, we are in particular also assuming a priori (6.15) for
some ε0 ≤ εslab , since this is included in the assumptions of Proposition 6.2.3. In view now of Corollary 4.5.3,
however, for α0 sufficiently large, if say ε̂slab (α) α−3 εslab , it follows from the additional assumptions (6.30)
that the estimates (6.15) in fact hold with the specific constant ε0 := ε̂0 α3 for all α ≥ α0 .
To obtain (6.34) we argue as follows. We note from the proof of Proposition 6.2.3 that we have the
following system of inequalities
(1) (1)
χ 1/2
X
k
. ε̂0 + ε0 χ X k
, (6.38)
(2 − δ) (2 − δ)
χ 1/2
X . ε̂0 + ε0 χk−1
k−1
X, (6.39)
q
(0) 1−δ 2δ (0) 1−δ 2δ 1/2(1) (0) (0) (1) (0)
χ
X . ε̂0 αL−1 + ( χk−1
k−1
X ) 1+δ ( χk−1
X ) 1+δ kX + ( k X ) 1+δ + ε0 χk−1
X ) 1+δ ( k X, (6.40)
(1) (1)
1/2
X . ε̂0 αL−1+δ + ε0 χk−2
χ
k−2
X, (6.41)
q
(0) 1−δ 2δ (0) 1−δ 2δ 1/2 (1) (0) (0) (1) (0)
χ 2 −2+δ χ 1+δ ( χ X ) 1+δ
X
k−3
. ε̂ 0 α L + ( X
k−3
) k−3
X + ( k
k
X ) 1+δ + ε0 χk−3
X ) 1+δ ( k X, (6.42)
where we have used also the initial data assumptions (6.30).
For ε̂slab (α) sufficiently small, we have ε0 1 and thus we obtain immediately that
(1) (2 − δ) (1)
χ
X
k
X . ε̂0 ,
. ε̂0 , χ
X . ε̂0 αL−1+δ ,
k−2
χ
k−1
q
(0) 1−δ 2δ (0) 1−δ 2δ (1) (0) (0) (1)
χ −1 χ 1+δ ( χ X ) 1+δ
X
k−1
. ε̂ 0 αL + ( X
k−1
) k−1
X + ( k
k
X ) 1+δ ( k
X ) 1+δ , (6.43)
q
(0) 1−δ 2δ (0) 1−δ 2δ (1) (0) (0) (1)
χ 2 −2+δ χ χ
X . ε̂0 α L
k−3
X)
+ ( k−3 1+δ X)
( k−3 1+δ X + ( k
k
X ) 1+δ ( k
X ) 1+δ .
60
The first two inequalities of (6.43) yield the first two inequalities of (6.34), while the third inequality of (6.43)
yields the fourth inequality of (6.34). On the other hand, from the latter inequality of (6.43) we obtain
(0)
1/2 3 (1−δ)(1+2δ)
χ
X . ε̂0 α2 L−2+δ + ε̂0 ε0 L− 2
k−3
1+δ . (6.44)
hence, plugging this again into the final inequality of (6.43) to estimate the nonlinear term, we obtain
(0)
1/2 3 (1−δ)(γ+2δ)
χ
X . ε̂0 α2 L−2+δ + ε̂0 ε0 L− 2
k−3
1+δ . (6.45)
This yields the fifth inequality of (6.34). Note that this implies
(0)
Finally, plugging these bounds into the fourth inequality of (6.43) yields the third inequality of (6.34),
completing the proof of (6.34).
To obtain (6.32), we recall (6.10) from Proposition 6.2.1. This gives, for sufficiently small ε̂slab ,
(1)
q (1)
q √ (1) (1) (0) (0)
χ χ
sup E(τ
k
) ≤ E(τ k
0) + C X k
X +C X
k k
X L,
k
τ0 ≤τ ≤τ0 +L
1/2 1/2
≤ ε̂0 + C ε̂0 ε0 L(−1+δ)/2 + C ε̂0 ε0 L(−2+δ)/2 L1/2
1/4
≤ ε̂0 (1 + ε̂0 L(−1+δ)/2 )
where we have used the first inequality of (6.30), the first inequality of (6.43), the estimate (6.47) and the
1/2 1/4
estimate (6.46) and we have replaced ε0 by ε̂0 in the penultimate line, sacrificing a quarter power to
absorb the resulting α term and the constant C.
To obtain (6.33), we now recall (6.9) from Proposition 6.2.1. This gives, for sufficiently small ε̂slab ,
(2 − δ)
q
(2 − δ)
q q q (2 − δ)
q √ (0) (2 − δ) (0) (2−δ) (0) (0)
where we have used the second inequality of (6.30), the second and third inequalities of (6.34), the esti-
1/2 1/4
mate (6.46) and the first inequality of (6.44), and we have replaced ε0 by ε̂0 in the penultimate line,
sacrificing a quarter power to absorb the resulting α term and the constant C.
To show (6.35)–(6.37), we first apply the pigeonhole principle as in [DR09] to the inequality
Z τ0 +L (0)
(1 − δ) (0)
0 0 0 0 −1 1−δ 0 0
E
k−1
(τ ) + E
k−2
(τ ) + α L E
k−3
(τ ) dτ 0 . ε̂0 ,
τ0
61
which, upon addition, follows from the first, second and fourth inequalities of the estimate (6.34) already
shown. Recalling from (3.92) that for both p = 2 − δ and p = 1, we have
(p−1) (p−1) (p−1) (p−1)
E 0 & k−2
k−2
E, E 0 & k−3
k−3
E,
we obtain that there exists τ 00 ∈ [τ0 , τ0 + L/2], whose precise value depends on the solution, such that
(0)
where we have used (6.49) and the estimate for k−2 E contained in (6.34).
Now we apply (6.10) and (6.11) again, with τ 00 in place of τ0 , using (6.48), (6.51) and (6.50) to bound
the initial data, to obtain that for all τ0 + L ≥ τ 0 ≥ τ0 + L/2, we have
(0) (0)
E(τ 0 ) ∼
k−1
E(τ 0 ) . ε̂0 L−1 ,
k−1
(6.52)
(1) (1)
E(τ 0 ) ∼
k−2
E(τ 0 ) . ε̂0 L−1+δ ,
k−2
(6.53)
(0) (0)
E(τ 0 ) ∼
k−3
E(τ 0 ) . ε̂0 αL−2+δ .
k−3
(6.54)
Thus, in view of the requirement α ≥ α0 1, for sufficiently large α0 we may absorb the constants
implicit in . by explicit constants of our choice by adding extra positive α powers to the right hand side
of (6.53) and (6.54). In this way, we obtain the specific estimates (6.36) and (6.37) for all τ 0 ≥ τ 00 and thus
in particular for all L ≥ τ 0 − τ0 ≥ L/2. In the same way, we also obtain the specific constant of the estimate
of (6.31) which will be convenient in our scheme.
Let us state an alternative “relaxed” version of the above proposition.
Proposition 6.2.6. Under the relaxed assumptions for Proposition 6.2.3 described in Remark 6.2.4, there
exists a constant C > 0, implicit in the inequalities . below, a parameter β > 0 sufficiently small, a parameter
α0 1 and, for all α ≥ α0 , a parameter ε̂slab (α) such that the following holds.
Given 0 < ε̂0 ≤ ε̂slab , let us assume in addition that
(1) (2 − δ) (0) (1) (0)
E(τ
k
β
0 ) ≤ ε̂0 L , E(τ0 ) ≤ ε̂0 Lβ ,
k−1
E(τ0 ) ≤ ε̂0 α1/2 L−1+β ,
k−1
E(τ0 ) ≤ ε̂0 α2 L−2+δ+β . k−2
E(τ0 ) ≤ ε̂0 αL−1+δ+β , k−3
(6.55)
Then the solution ψ of (4.1) on R(τ0 , τ0 + L) given by Proposition 6.2.3 satisfies the additional estimates
(1) (2 − δ)
sup E(τ
k
) ≤ αβ ε̂0 Lβ , sup E(τ ) ≤ αβ ε̂0 Lβ ,
k−1
(6.56)
τ ≤τ ≤τ0 +L τ0 ≤τ ≤τ0 +L
(1) (2 − δ) (0) (1) (0)
χ
X
k
. ε̂0 Lβ , χ
X . ε̂0 Lβ ,
k−1
χ
k−1
X . ε̂0 αL−1+β , χ
X . ε̂0 αL−1+δ+β ,
k−2
χ
k−3
X . ε̂0 α2 L−2+δ+β . (6.57)
Moreover, for all times τ 0 with L ≥ τ 0 − τ0 ≥ L/2, we have that
(0) (1) (0)
62
6.2.4 The iteration: proof of Theorem 5.1 in case (ii)
We may now prove Theorem 5.1 in case (ii).
We shall proceed iteratively.
We define τ0 = 1, L0 = 1, Li = 2i , τi+1 = τi +Li , and fix α ≥ α0 so that the statement of Proposition 6.2.5
holds. (Note that we shall no longer track the dependence of constants and parameters on α since it is now
fixed; one could simply take α = α0 .)
For 0 < ε0 ≤ εglobal and εglobal sufficiently small, since
(1) (2 − δ)
E(τ
k
E(τ0 ) ≤ ε0 ,
0 ) +k−1
E(τ
k
0 ) . ε0 , E(τ0 ) . ε0 .
k−1
Thus, for sufficiently small εglobal ε̂slab , it follows that there exists a ε̂0 (ε0 ) ∼ ε0 , satisfying moreover
2αε̂0 ≤ εslab , such that
1
ε̂0 ≤ ε̂slab
2
and
(1) (2 − δ) (0) (1) (0)
E(τ
k
0 ) ≤ ε̂0 , E(τ0 ) ≤ ε̂0 ,
k−1
E(τ0 ) ≤ ε̂0 αL−1 ,
k−1
E(τ0 ) ≤ ε̂0 αL0−1+δ ,
k−2
E(τ0 ) ≤ ε̂0 α2 L−2+δ .
k−3
E(τ
k
0 ) ≤ αε̂0 .
In general, given τi ≥ 1 defined above, ε̂i ≤ ε̂slab , and a solution ψ of (4.1) on R(τ0 , τi ) such that
(1)
E(τ
k
i ) ≤ αε̂i (6.59)
and
(1) (2 − δ) (0) (1) (0)
E(τ
k
i ) ≤ ε̂i , E(τi ) ≤ ε̂i ,
k−1
E(τi ) ≤ ε̂i αL−1 ,
k−1
E(τi ) ≤ ε̂i αLi−1+δ ,
k−1
E(τi ) ≤ ε̂i α2 Li−2+δ ,
k−2
(6.60)
note that by our restrictions on ε̂slab , the assumptions of Proposition 6.2.3 hold with α2 ε̂i in place of ε0 (here
we are using (6.59) to invoke Corollary 4.5.3 to rewrite (6.60) in terms of the calligraphic energies; cf. the
first lines of the proof of Proposition 6.2.5) and the assumptions of Proposition 6.2.5 then apply with ε̂i in
place of ε̂0 , where both propositions are understood now with τi , τi+1 in place of τ0 , τ1 . It follows that the
solution ψ extends to a solution defined also in Ri := R(τi , τi + Li ) satisfying the estimates (6.31)–(6.34)
while for τ 0 = τi+1 = τi + Li we have in addition (6.36)–(6.37). We have thus
(1)
E(τ
k
i+1 ) ≤ αε̂i ≤αε̂i+1 ,
(1)
1/4 −(1+δ)/2
E(τ
k
i+1 ) ≤ ε̂i (1 + ε̂i Li ) ≤ε̂i+1
(2 − δ)
1/4 −1/2
E(τi+1 ) ≤ ε̂i (1 +
k−1
ε̂i Li ) ≤ε̂i+1 ,
(1)
1
E(τi+1 ) ≤ ε̂i αL−1 i ≤ε̂i+1 αL−1
i+1 ,
k−1
2
(1)
1
E(τi+1 ) ≤ ε̂i αL−1+δ
i
−1+δ
≤ε̂i+1 αLi+1 ,
k−2
2
(0)
1
E(τi+1 ) ≤ ε̂i α2 L−2+δ
i
−2+δ
≤ε̂i+1 α2 Li+1 ,
k−3
4
as long as
1/4 −(1+δ)/2
ε̂i+1 := ε̂i (1 + ε̂i Li ). (6.61)
63
Now note that for εglobal sufficiently small, the above inductive definition (6.61) of ε̂i ensures that ε̂i ≤ 2ε̂0
for all i.
It follows that a solution exists in R(τ0 , ∞) = ∪R(τi , τi + Li ) and in each interval the estimates (6.34)
hold.
We obtain finally that for all τ ≥ 1 we have
(1) (2 − δ)
3/2 3/2
E(τ
k
) ≤ ε̂0 + C ε̂0 , E(τ ) ≤ ε̂0 + C ε̂0 ,
k−1
(6.62)
Z τ (1 − δ) (1 − δ)
χ 0
k−1
E 0 . ε̂0 log(τ + 1),
E +k−2 (6.63)
τ0
(1) (1)
To show (6.68), for all τ ≥ τ0 , we reapply the estimates of Proposition 4.8.1 globally on R(τ0 , τ ). To control
the nonlinear error bulk integrals, redecompose the domain in dyadic time slabs and apply the estimates
3/2
proven and sum. In view of the estimates, all these error bulk integrals can be controlled by C ε̂0 .
6.2.5 Alternative proof using Proposition 6.2.6 and the semilinear case
We note that we may prove the theorem alternatively using Proposition 6.2.6. We first fix α ≥ α0 , and now
define Li := αi . Here our iterative assumption is
(1) (2 − δ) (0) (1) (0)
E(τ
k
β
0 ) ≤ ε̂0 Li , E(τ0 ) ≤ ε̂0 Lβi ,
k−1
E(τ0 ) ≤ ε̂0 αLi−1+β ,
k−1 k−2
E(τ0 ) ≤ ε̂0 αLi−1+δ+β , E(τ0 ) ≤ ε̂0 α2 L−2+δ+β
k−3 i .
In view of (6.56) and (6.58) and the definition of Li , we see that this is closed under iteration. We obtain
thus existence in R(τ0 , ∞), and at first instance the bounds
(1) (2 − δ)
E(τ
k
) . ε̂0 τ β , k−1
E(τ ) . ε̂0 τ β ,
(0) (1) (0)
These bounds are of course weaker than those of (6.62)–(6.67). The τ β factors terms may be, however,
removed a posteriori, as in the last paragraph of Section 6.2.4, by revisiting the estimates globally on R(τ0 , τ ),
controlling the nonlinear error bulk integrals by redecomposing into dyadic time slabs, applying the estimates
already proven, and summing.
Thus this proof in the end yields finally the same estimates as before, but where (6.62) is replaced by
(0) (1)
Remark 6.2.8. In view of Remark 6.2.7, this proof in particular applies to the semilinear case with the
relaxed assumptions of Remark 4.8.4, giving the result for case (ii) as stated in Remark 5.1.
Remark 6.2.9. The disadvantage of this alternative proof over the proof given in Section 6.2.4 is that to
obtain the fundamental top-order orbital stability statement (6.69), one must revisit the estimates globally.
We thus believe that the proof in Section 6.2.4 better reflects the dyadically localised philosophy of our method.
64
6.3 Case (iii)
We now turn to the most general case we consider, case (iii), where we do not assume that (3.3) follows from
a physical space identity (3.14), but only assume (3.3) as a black box, together with the weaker physical
space identity described in Section 3.4.3.
This case is slightly more involved because we must combine the estimate originating from Section 3.4.3
with the estimate originating in Section 3.2. As we shall see, however, the difference with case (ii) is in fact
quite minor.
(2 − δ) (2 − δ)
q q q q(2 − δ) (0)
q √ (2 − δ) (0) (2−δ) (0) (0)
χ ρ ρX ρX ρ
X
k−1
. E(τ
k−1
0 ) + X
k−1
X
k
+ k−1 k−1
X
k
+ X
k
X L,
k
(6.71)
(1) (1) (1) (1)
q (0)
q √ (1) (1) (0) (0)
F(v, τ1 ), E(τ
k
1 ), c ρXk
≤ E(τ
k
0 ) + AC k−1
χ
X + C ρX k
X + C ρX
k k
X L,
k
(6.72)
k
(1) (1)
q q √ (1) (0) (0) (0)
χ
k−1
E(τ0 ) + C ρk−1
X . k−1 X k X + C ρX k
X L,
k
(6.73)
q
(0) (0) (0) (0) 1−δ
(0)
2δ (0)1−δ 2δ (0) (1) (0) (0) (1)
F(v, τ1 ), E(τ
k
1 ), c ρXk
≤ E(τ
k
0 ) + AC k−1
χ
X + C ρX k
+ ( ρXk
) 1+δ ( ρ Xk
) 1+δ X + ( k
k
X ) 1+δ ( k
X ) 1+δ
k
q √ (0) (0)
ρ
+C X k
X L,
k
(6.74)
(0)
(0) 1−δ (0)
2δ
q (0) 1−δ 2δ (1) (0) (0) (1)
χ
k−1
E(τ0 ) + ρk−1
X . k−1 X + ( ρk−1
X ) 1+δ ( ρk−1
X ) 1+δ X + ( k
k
X ) 1+δ ( k
X ) 1+δ
q √ (0) (0)
+ρ Xk
X L.
k
(6.75)
Proof. Again we recall that if εboot ≤ εloc , then the assumption of Proposition 4.8.1 holds. The proposition
then follows from Proposition 4.8.1 as in the proof of Proposition 6.2.1, where we have used also the crucial
estimate Z τ1 ξ (0)
A E 0 (τ 0 )dτ 0 . AC χk−1
k−1
X,
τ0
which follows from the fundamental relation (3.99).
E(τ
k
0 ) ≤ ε0 , E(τ0 ) ≤ ε0 L−1 ,
k−2
(6.76)
then the unique solution of Proposition 4.9.1 obtaining the data can be extended to a ψ defined on the entire
spacetime slab R(τ0 , τ0 + L) satisfying the equation (4.1) and the estimates
(1) (1) (0) (0)
ρ
X
k
+ χk−1
X . ε0 , ρ
k−2
X . ε0 L−1 .
X + χk−3 (6.77)
65
Proof. Consider the set B ⊂ (τ0 , τ0 + L] consisting of all τ0 + L ≥ τf ≥ τ0 such that a solution ψ of (4.1)
obtaining the data exists on R(τ0 , τf ) and such that the boostrap assumption (4.23) (with p = 1) and also
the additional bootstrap assumption
(0)
X ≤ εL−1
k
(6.78)
hold in R(τ0 , τ1 := τf ), where 0 < ε ≤ εboot is a small constant satisfying
1 ε εslab . (6.79)
E(τ
k
0 ) ≤ ε̂0 , E(τ0 ) ≤ ε̂0 ,
k−2
E(τ0 ) ≤ ε̂0 αL−1 ,
k−2
E(τ0 ) ≤ ε̂0 αL−1+δ ,
k−4
E(τ0 ) ≤ ε̂0 α2 L−2+δ . (6.85)
k−6
Then the solution ψ of (4.1) on R(τ0 , τ0 + L) given by Proposition 6.2.3 satisfies the additional estimates
(1)
sup E(τ
k
) ≤ αε̂0 , (6.86)
τ0 ≤τ ≤τ0 +L
66
(1)
sup E(τ
k
) ≤ ε̂0 (1 + αL−1/4 ), (6.87)
τ0 ≤τ ≤τ0 +L
(2 − δ)
(0) (0)
(0) 1−δ (1)
2δ
q (0) (0) 1−δ (1)
2δ
ρ
k−2
X . ε̂0 αL−1 +
X + χk−3 ρ
X ) 1+δ ( ρk−2
k−2
X ) 1+δ X + ( k
k
X ) 1+δ ( k
X ) 1+δ , (6.95)
(0) (0)
1/2 3 (1−δ)(γ+2δ)
ρ
k−6
X . ε̂0 α2 L−2+δ + ε̂0 ε0 L− 2
X + χk−7 1+δ . (6.96)
The inequalities (6.94) give the first second and fourth inequality of (6.89).
Note that −2 + δ < − 23 + δ. We may however improve (6.96) iteratively as follows: If
(0) (0)
ρ
X . ε̂0 α2 L−γ ,
X + χk−7
k−6
for γ ≤ 2 − δ, then
(0)
k
X . ε̂0 α2 L−γ . ε0 L−γ ,
hence, plugging this again into (6.81), we obtain
(0) (0)
1/2 3 (1−δ)(γ+2δ)
ρ
k−6
X . ε̂0 α2 L−2+δ + ε̂0 ε0 L− 2
X + χk−7 1+δ . (6.97)
3 (1−δ)(1+2δ) 3 (1−δ)(γi−1 +2δ)
Setting γ0 = 2 1+δ , and defining inductively γi = 2 1+δ , we have that there exists a first
3 (1−δ)(γi +2δ)
i ≥ 1 such that 2 1+δ > 2 − δ. It follows that (6.97) holds for γ = γi , hence
(0) (0)
ρ
k−6
X . ε̂0 α2 L−2+δ .
X + χk−7
67
We may now infer the third inequality of (6.89) by plugging in the derived bounds into (6.95). This concludes
the proof of (6.89).
To show (6.91)–(6.93), respectively, we first apply the pigeonhole principle as in [DR09] to the inequality
Z τ0 +L (0) (1 − δ)
(0)
E 0 (τ 0 ) +k−4
k−2
E 0 (τ 0 ) + α−1 L1−δk−6
E 0 (τ 0 ) dτ 0 . ε̂0 ,
τ0
which, upon addition, follows from the inequalities of the estimate (6.89) already shown. Recalling from (3.92)
that we have
(0) (0) (1 − δ) (1 − δ) (0) (0)
E 0 & k−2
k−2
E, E 0 & k−4
k−4
E, E 0 & k−6
k−6
E,
we obtain that there exists τ 00 ∈ [τ0 , τ0 + L/2], whose precise value depends on the solution, such that
(0)
k−2
E(τ 00 ) . ε̂0 · L−1 , (6.100)
(1 − δ)
k−4
E(τ 00 ) . ε̂0 · L−1 , (6.101)
(0)
k−6
E(τ 00 ) . ε̂0 αL−1+δ · L−1 . (6.102)
E(τ 00 ) .
k−4
E(τ 00 )
k−4
E(τ 00 )
k−4
≤ E(τ 00 )
k−4
sup E(τ
k−4
) (6.103)
τ0 ≤τ ≤τ0 +L
E(τ 0 ) ∼ k−2
k−2
E(τ ) . ε̂0 L−1 , (6.105)
(1) (1)
E(τ 0 ) ∼ k−4
k−4
E(τ ) . ε̂0 L−1+δ , (6.106)
(0) (0)
E(τ 0 ) ∼ k−6
k−6
E(τ ) . ε̂0 αL−2+δ . (6.107)
Thus, in view of the requirement α ≥ α0 1, for sufficiently large α0 we may absorb the constants
implicit in . by explicit constants of our choice by adding extra positive α powers to the right hand side
of (6.106) and (6.107). In this way, we obtain the specific estimates (6.91), (6.92) and (6.93). In the same
way, we also obtain the specific constant of the estimate of (6.86) which will be convenient in our scheme.
We finally turn to (6.87) and (6.88). Let us first note that revisiting (6.75), with a little bit of averaging,
we have the bounds
Z τ0 + 12 q √
χ
(0)
1 (0)
ρ ρ 1−δ 2δ
1+δ ( ρ X ) 1+δ
q (0) 1−δ (0)
2δ
1+δ ( X ) 1+δ + ρ X
(1) (0) (0) (1) (0) (0)
s s
Z τ0 + 21 Z τ0 + 12
(0) (0)
1 1
. E(τ
k
)dτ E(τ )dτ + ε̂0 ε 2 L− 4
k−2
τ0 τ0
1
− 12 1
− 41
. ε̂0 α L 2 + ε̂0 ε L 2
1
− 14
. ε̂0 α L 2 ,
where we have used also the interpolation inequality (3.114) (and that L ≥ 12 ).
68
(0) (0) R τ0 + 12 (0)
X 0 := χk−1
We now notice that the above bound clearly holds also for χk−1 X (τ0 + 12 , τ0 + L) + τ0
E(τ )dτ ,
k−1
(0) (0)
For (6.87), we revisit (6.72), noting that we may replace χk−1 X 0 on the right hand side, and thus we
X by χk−1
may bound
(1)
q(1)
q √ (0) (1) (1) (0) (0)
χ 0 ρ ρ
sup E(τ
k
) ≤ E(τ
k
0 ) + AC X
k−1
+ C X
k
X
k
+ C X
k
X L
k
τ ∈[τ0 ,τ0 +L]
1 1 1/2 3
≤ ε̂0 + AC(ε̂0 α 2 L− 4 ) + C ε̂0 ε0 L(−1+δ)/2 + C ε̂0 ε12
0 L
− 2 +δ
1
≤ ε̂0 + αε̂0 L− 4 ,
. ε̂0 αL−1
whence we have
(2 − δ) (2 − δ) (0)
q
(2 − δ)
q q q (0)
q √ (2 − δ) (0) (2−δ) (0) (0)
χ ρ ρ ρ ρ
sup E(τ ) ≤ k−1
k−1
X +C
E(τ0 ) + A k−2 X k
k−1
X+ X k−1
k−1
X k
X + C k−1
X kX L
τ ∈[τ0 ,τ0 +L]
1 1/2 1/2
≤ ε̂0 + Aε̂0 α 2 L−1 + C ε̂0 ε0 L(−2+δ)/2 + C ε̂0 ε0 L(−2+δ)/2 L1/2
1
≤ ε̂0 + αε̂0 L− 4 ,
where we have again used a higher power of α to absorb all other constants.
(Note that we shall no longer note the dependence of constants and parameters on α since it is now
considered fixed. Thus implicit constants depending on the choice of α will from now on be incorporated in
the notations ∼ and ..)
For 0 < ε0 ≤ εglobal and εglobal sufficiently small, since
(1) (2 − δ)
E(τ
k
0 ) ≤ ε0 , E(τ0 ) ≤ ε0 ,
k−2
(6.109)
E(τ
k
0 ) . ε0 , E(τ0 ) . ε0 .
k−2
Thus, for sufficiently small εglobal ε̂slab , and all 0 < ε0 ≤ εglobal , then if (6.109) is satisfied, it follows that
there exists a ε̂0 (ε0 ) ∼ ε0 , satisfying
1 1
αε̂0 ≤ εslab , ε̂0 ≤ ε̂slab (6.110)
d d
and
(1) (2 − δ) (0) (1) (0)
E(τ
k
0 ) ≤ ε̂0 , E(τ0 ) ≤ ε̂0 ,
k−2
E(τ0 ) ≤ ε̂0 αL−1
k−2 0 , E(τ0 ) ≤ ε̂0 αL0−1+δ ,
k−4
E(τ0 ) ≤ ε̂0 α2 L−2+δ
k−6 0 .
69
Finally, with our restriction on the definition of ε0 , we may also write
(1)
E(τ
k
0 ) ≤ αε̂0 .
In general, given τi ≥ 1 defined above, ε̂i ≤ ε̂slab , and a solution ψ of (4.1) on R(τ0 , τi ) such that
(1)
E(τ
k
i ) ≤ αε̂i , (6.111)
and
(1) (2 − δ) (0) (1) (0)
E(τ
k
i ) ≤ ε̂i , E(τi ) ≤ ε̂i ,
k−2
E(τi ) ≤ ε̂i αL−1
k−2 i , E(τi ) ≤ ε̂i αL−1+δ
k−4 i , E(τi ) ≤ ε̂i α2 Li−2+δ ,
k−6
(6.112)
note that by our restrictions on ε̂slab , the assumptions of Proposition 6.2.3 hold with α2 ε̂i in place of ε0 (here
we are using (6.111) to invoke Corollary 4.5.3 to rewrite (6.60) in terms of the calligraphic energies; cf. the
first lines of the proof of Proposition 6.3.3) and the assumptions of Proposition 6.3.3 then apply with ε̂i in
place of ε̂0 , where both propositions are understood now with τi , τi+1 in place of τ0 , τ1 . It follows that the
solution ψ extends to a solution defined also in Ri := R(τi , τi + Li ) satisfying the estimates (6.86)–(6.89)
while for τ 0 = τi+1 = τi + Li we have in addition (6.92)–(6.93). We have thus
(1)
E(τ
k
i+1 ) ≤ ε̂i α ≤ε̂i+1 α,
(1)
− 14
E(τ
k
i+1 ) ≤ ε̂i (1 + αLi ) ≤ε̂i+1 ,
(2 − δ)
− 14
E(τi+1 ) ≤ ε̂i (1 + αLi )
k−2
≤ε̂i+1 ,
(1)
1
E(τi+1 ) ≤ ε̂i αL−1 i ≤ε̂i+1 αL−1
i+1 ,
k−2
2
(1)
1
E(τi+1 ) ≤ ε̂i αL−1+δ
i
−1+δ
≤ε̂i+1 αLi+1 ,
k−4
2
(0)
1
E(τi+1 ) ≤ ε̂i α2 L−2+δ
i ≤ε̂i+1 α2 L−2+δ
i+1 ,
k−6
4
as long as
−1
ε̂i+1 := ε̂i (1 + αLi 4 ). (6.113)
By our requirement (6.110) and the definition (6.108) of the parameter d, then defining ε̂i+1 inductively
by (6.113), it follows that ε̂i+1 ≤ ε̂slab and αε̂i+1 ≤ εslab .
It follows that a solution exists in R(τ0 , ∞) = ∪R(τi , τi + Li ) and in each interval the estimates (6.89)–
(6.90) hold, with L = Li .
We obtain finally that for all τ ≥ 1 we have (among other estimates)
(1)
E(τ
k
) . ε0 , (6.114)
Z τ (0) (0) (0)
ρ 0
E 0 +k−2
Ek + χk−1 E 0 . ε0 log(τ + 1), (6.115)
τ0
(2 − δ)
E(τ ) . ε0 ,
k−2
(6.116)
(1) (0)
E(τ ) . ε0 τ −1+δ ,
k−4
F(v, τ ) . ε0 τ −1+δ ,
k−4
(6.117)
Z ∞ (0) (0) (0)
ρ 0
E 0 +k−6
E + χk−5
k−4
E 0 . ε0 τ −1+δ , (6.118)
τ
(0) (0)
E(τ ) . ε0 τ −2+δ ,
k−6
F(v, τ ) . ε0 τ −2+δ ,
k−6
(6.119)
Z ∞ (−1−δ) (−1−δ) (−1−δ)
ρ 0
E + χk−7
k−6
E 0 . ε0 τ −2+δ .
E 0 +k−8 (6.120)
τ
70
Finally, let us note that we may improve (6.115) to
Z ∞ (0) (0) (0)
ρ 0
E 0 +k−2
Ek + χk−1 E 0 . ε0 . (6.121)
τ0
This follows, for all τ , by applying again both estimates of Proposition 4.8.1 appropriate to case (iii) globally
in R(τ0 , τ ). One may now re-estimate all nonlinear spacetime integrals arising in the estimates on dyadic
intervals and sum.
∆ = r2 − 2M r + a2 , Σ = r2 + a2 cos2 θ . (A.1)
We recall the (t, r, θ, φ) Boyer–Lindquist coordinates and the associated (t, r? , θ, φ) coordinates (which we
will also refer to as Boyer–Lindquist coordinates) with the familiar relation
dr? r2 + a2
= .
dr ∆
We shall consider these in the domain R × (r+ , ∞) × S2 , where (θ, φ) are identified with the usual spherical
coordinates of S2 . The Kerr metric then has the form
g = ga,M = gtt dt2 + gtφ dtdφ + gr? r? (dr? )2 + gθθ dθ2 + gφφ dφ2 ,
71
with
2M r Σ∆
gtt = − 1 − , gr? r? = 2 , gθθ = Σ ,
Σ (r + a2 )2
2M ra2
2M r
gtφ = − a sin2 θ , gφφ = r2 + a2 + sin2 θ sin2 θ . (A.2)
Σ Σ
2M ra2 (r2 + a2 )2
1 ? ? 1
g tt = − r2 + a2 + sin2 θ , gr r
= , g θθ = ,
∆ Σ Σ∆ Σ
2M r ∆ − a2 sin2 θ
g tφ = − a , g φφ = . (A.3)
Σ∆ Σ∆ sin2 θ
For the determinant in these coordinates we note
√ ∆
Σ sin θ = gr? r? r2 + a2 sin θ .
g= (A.4)
r2 +a 2
where
g β 1
∂r? r2 + a2 ∂r? β
V=− =− 2 2
β gr? r? (r + a )β
1 2M r + a2 r(r − 4M ) + a4
3
1
= · =: V0 . (A.7)
Σ (r2 + a2 )2 Σ
˜ ν ψ − 1 gµν ∇
˜ µ ψ∇
T̃µν [ψ] = ∇ ˜ µ ψ∇
˜ µ ψ + Vψ 2 (A.8)
2
1
= β ∇µ (β ψ)∇ν (β −1 ψ) − gµν ∇µ (β −1 ψ)∇µ (β −1 ψ) + Vβ −2 ψ 2 .
2 −1
2
We note that for all |a| < M , V is strictly positive in the exterior, in fact
2M r(r − M )(r + M )
V≥ & r−3 , (A.9)
Σ(r2 + a2 )2
since r ≥ r+ > M , and T̃µν [ψ] satisfies the dominant energy condition, i.e. T̃µν [ψ]ξ µ ξ ν is nonnegative if ξ is
timelike, and in fact controls coercively first derivatives of ψ as well as ψ itself (the latter with the weight
r−3 ). (Indeed, this direct control of the zero’th order term is our motivation for considering (A.8) in place
of the usual Tµν .).
From Proposition 3 of [HW14] we infer:
72
Proposition A.2.1. For ψ a C 2 solution of g ψ = 0 and X a smooth spacetime vector field we have the
identity
where
and
˜ †µ (βV)
∇ ˜†β
∇ µ ˜ν ˜
S̃µ [ψ] = ψ2 + ∇ ψ ∇ν ψ . (A.13)
2β 2β
We also have the Lagrangian identity for an arbitrary spacetime function w:
Proposition A.2.2. For ψ a C 2 solution of g ψ = 0 and w a smooth spacetime function we have the
identity
with
1
K̃ aux,w [ψ] := wβ 2 ∇α (β −1 ψ)∇α (β −1 ψ) + (β −1 ψ)2 − ∇µ β 2 ∇µ w + Vwβ 2 ,
2
1
J˜µaux,w := wβ 2 β −1 ψ∇µ (β −1 ψ) − ψ 2 ∇µ w.
2
Proof. This follows from
˜†∇
∇µ w ψβ −1 q 2 ∇µ (β −1 ψ) =wβ 2 ∇µ (β −1 ψ)∇µ (β −1 ψ) + wψ −∇ ˜ µψ
µ
1
+ (∇µ w)β 2 ∇µ (β −1 ψ)2 (A.15)
2
after rearranging and inserting (A.6).
Remark A.2.3. Note that when w is a function of r only (as will always be the case in the applications
below) we have
1 √ ? ? r2 + a2
∇µ β 2 ∇µ w = √ ∂r? gg r r β 2 ∂r? w = ∂ 2? w ,
g ∆Σ r
a formula which is useful in the computations below.
73
√
Lemma A.3.1. For β −1 = r2 + a2 , h = 1 − 3M ∆
(r 2 +a2 )2 and X = f (r)∂r (with f arbitrary) we have
?
r
for any δ ∈ R the divergence identity
aux, 1 f 0
∇µ J˜X [ψ] + J˜µ 2 [ψ] − δ · J˜aux,f ·h [ψ]
µ µ
aux, 12 f 0
X
= K̃ [ψ] + K̃ [ψ] − δ · K̃µaux,f ·h [ψ] , (A.16)
where
1 0
aux, f
J˜µX [ψ] + J˜µ 2 [ψ] − δJ˜µaux,f ·h [ψ]
1
= T̃µν [ψ]X ν + f 0 β 2 β −1 ψ∇µ (β −1 ψ) − ψ 2 ∇µ f 0
2
2 −1 −1
1 2
− δ f · hβ β ψ∇µ (β ψ) − ψ ∇µ f · h (A.17)
2
and
1 0
K̃ f ∂r? [ψ] + K̃ aux, 2 f [ψ] − δK̃µaux,f ·h [ψ]
? ? 2
= β 2 g r r ∂r? f − δf h ∂r? (β −1 ψ)
1 X
+ f β2 (Aµν − 2δhg µν ) ∂µ (β −1 ψ)∂ν (β −1 ψ)
2
µ,ν6=r ?
r + a2
2
1 1 ∆
+ − f 000 − f ∂r? V0
∆Σ 4 2 (r2 + a2 )2
!
1 00 ∆
+δ (f · h) − V0 f · h 2 (q −1 ψ)2 (A.18)
2 (r + a2 )2
with
h ? ?
i
Aµν = −∂r? (g µν ) + g µν gr? r? ∂r? (g r r ) . (A.19)
1 X h ? ?
i
− f β2 ∂r? (g µν ) − g µν gr? r? ∂r? (g r r ) ∂µ (β −1 ψ)∂ν (β −1 ψ)
2 ? µ,ν6=r
1 1
− β 2 (∂r? f ) g αβ ∂α (q −1 ψ)∂β (β −1 ψ) − f ∂r? (β 2 V)(β −1 ψ)2
2 2
1 1 √ 2
− ∂r? f + f √ ∂r? g Vψ . (A.20)
2 g
Adding the Lagrangian identity of Proposition A.2.2 with w = 12 f 0 (and using Remark A.2.3) we deduce
the result for δ = 0. For arbitrary δ we simply add the Lagrangian identity of Proposition A.2.2 with
w = 1 − 3M r
∆
(r 2 +a2 ) f = f · h and group terms.
We now exploit the divergence identity of Lemma A.3.1. For this we first define the following (disjoint)
decomposition of the range of the r-variable:
[r+ , ∞) = R1 ∪ R2 ∪ R3 ∪ R4 ∪ R5
5 5 11 11 7 7
= r+ , M ∪ M, M ∪ M, M ∪ M, 4M ∪ [4M, ∞) .
2 2 4 4 2 2
a
Note that M ∩ {r ∈ R3 } includes the region containing all trapped null geodesics if M is suitably small.
74
|a|
Proposition A.3.2. There exists an (explicit) function f such that for all M sufficiently small we can
choose δ > 0 sufficiently small (depending only on M ) such that the estimate
f∆
1 0 aux, 2 2 2
K̃ f ∂r? [ψ] + K̃ aux, 2 f [ψ] − δ · K̃µ (r +a ) [ψ] ≥
2 2 !
∂t (β −1 ψ) + ∂r? (β −1 ψ) + (β −1 ψ)2 1 ˚ −1 2
c1R1 ∪R5 + 5 |∇(β ψ)|
r4 r
− C 1R2 ∪R4 |(β −1 ψ)|2 (A.21)
˚ −1 ψ)|2 :=
holds for constants c and C depending only on M . Here we have defined the shorthand |∇(β
−1
2 1 −1
2
∂θ (β ψ) + sin2 θ ∂φ (β ψ) .
Proof. In the proof, we let c̃ and C̃ be constants depending onlyon M (but which might change from line
to line). Starting from (A.18) we define B µν = Aµν − 2δ 1 − 3Mr
∆
(r 2 +a2 )2 g
µν
and compute
2
r3 − 3M r2 + a2 r + a2 M 3M r2 + a2 + 2MΣra sin2 θ
tt 2 2
B = 2a · sin θ + 2δ 1 − ,
Σ(r2 + a2 )2 r (r2 + a2 )2
2aM (a2 − 3r2 )
3M 2M r
B tφ = + 2δ · a 1 − ,
Σ(r2 + a2 )2 r Σ(r2 + a2 )2
r2 (r − 3M ) + a2 r cos[2θ] + a2 M ∆ − a2 sin2 θ
φφ 3M
B =2 − 2δ 1 − ,
Σ(r2 + a2 )2 sin2 θ r Σ(r2 + a2 )2 sin2 θ
r2 (r − 3M ) + a2 r + a2 M
3M ∆
B θθ = 2 2 − 2δ 1 − . (A.22)
2 2 2
Σ(r + a ) sin θ r Σ(r + a2 )2
2
B tt c̃ B φφ c̃ B θθ c̃
≥ 2 , ≥ 3 2 , ≥ 3 (A.23)
(1 − 3M r ) r (1 − 3Mr ) r sin θ (1 − 3Mr ) r
B tφ |a| C̃
≤ (A.24)
(1 − 3Mr ) M r4
for a fixed constant c̃ > 0 depending only on M . In particular (potentially making δ slightly smaller), we
can achieve that
? ? 3M ∆ 3M c̃
g r r ∂r? f − δf 1 − ≥ f 1 −
r (r2 + a2 )2 r r2
/ R3 } for a c̃ depending only on M . Since f 1 − 3M
holds in all of M ∩ {r ∈ r is globally non-negative and
in fact bounded uniformly below by a c̃ in R1 ∪ R5 , the estimate (A.21) now follows except for the zeroth
75
order term on the right hand side. (For this the only thing to notice is that in
µν
X
1 2 3M B
fβ 1 − ∂ (β −1 ψ)∂ν (β −1 ψ)
3M µ
2 r 1 − r
µ,ν6=r ?
the quadratic form in the square bracket is positive definite by the estimates (A.23) and (A.24).)
For the zeroth order term, clearly we only need to establish a lower bound in R1 ∪ R5 . We start with R5 ,
where we have for f = 1 − M r that
0
1 ∆ 3M
− f 000 − f V0 = + O(r−5 )
2 (r2 + a2 )2 r4
and 0 !
r2 + a2 M2
1 ∆ M 2M
− f 000 − f V0 = 4 3− − 14 2 if a = 0.
r2 − 2M r + a2 2 (r + a2 )2
2 r r r
Since the left hand side of the second identity is continuous in a and the bracket on the right hand side
uniformly bounded below by 13 for r ≥ 3M , the estimate claimed for theh zeroth order term followsifor
|a| 2 2 00
r +a 1 ∆
sufficiently small M after potentially making δ smaller (note δ r2 −2M r+a2 2 (f · h) − V0 f · h (r2 +a 2 )2 ≤
C̃δ
r 4 ).
Similarly, in R1 we have for f = − M
r the identity
r2 + a2
1 000 0 M
− f − f (wV0 ) = − 6 14M 2 − 14M r + 3r2
if a = 0.
r2 − 2M r + a2 2 r
3
The expression on the right hand side is easily shown to be uniformly bounded below by M 4r 6 for r ∈
[9M/5, 11M/4]. Since the expression on the left is in particular continuous in a on [r+ , 11M/4] the estimate
claimed for the zeroth order term also follows in R1 .
A.4 Adding the redshift and stationary currents and completing the proof
Let T denote the stationary Killing field of Kerr (which in Boyer–Lindquist coordinates is T = ∂t ) and N
the timelike (including on H+ ) redshift vector field constructed in Theorem 7.1 of [DR13]. We have that
K̃ T = 0 while
2 2
˚ −1 ψ)|2 + (β −1 ψ)2
K̃ N [ψ] ≥ c̃ ∂t (β −1 ψ) + ∂r? (β −1 ψ) + |∇(β
2 2
r + a2
a −1 9
+ ∂t − ∂r? + 2 2
∂φ (β ψ) in r ≤ M
∆ r +a 4
It is now clear that we can choose η sufficiently small (depending only on M ) such that for r ≥ 94 M we
1 0
can absorb the term η K̃ N [ψ], which is supported only in R1 , by the positivity of K̃ X [ψ] + K̃ aux, 2 f [ψ] − δ ·
76
K̃µaux,f ·h [ψ] established in Proposition A.3.2 to deduce
f∆
1 0 aux, 2 2 2
K̃ f ∂r? [ψ] + K̃ aux, 2 f [ψ] − δK̃µ (r +a ) [ψ] + η K̃ N [ψ] ≥ (A.27)
2 2 !
∂t (β −1 ψ) + ∂r? (β −1 ψ) + (β −1 ψ)2 1 ˚ −1 2
+ c1R1 ∪R5 + 5 |∇(β ψ)|
r4 r
2 2
r + a2
a
+ c1R1 ∂t − ∂r? + 2 ∂ φ (β −1
ψ) − C 1R2 ∪R4 |(β −1 ψ)|2 ,
∆ r + a2
where c may be smaller than in (A.21) but still depends only on M . We finally claim that we can choose Υ
sufficiently large in (A.26) such that
aux, 1 f 0
J˜µX [ψ] + J˜µ 2 [ψ] − δJ˜µaux,f ·h [ψ] + ΥJ˜µT [ψ] + η J˜µN [ψ] nµ Στ
For the boxed term, the estimate is an immediate consequence of the twisted energy momentum tensor
(A.8) satisfying the dominant energy condition, as discussed in Section A.2, the positivity (A.9), and the
fact that the vector field ΥT +ηN +X is timelike for sufficiently large Υ, provided that we then restrict to |a|
sufficiently small. Moreover, for the boxed term the zeroth order term in the estimate scales with Υ, i.e. the
larger we choose Υ the larger the zeroth order term becomes. This can be used to absorb the remaining
terms using the Cauchy–Schwarz inequality, for |a| sufficiently small depending on this final choice of Υ.
The relations (A.27) and (A.28) give the coercivity properties (3.30) and the first of (3.31), restricted
to the exterior region, where we define ρ = 1R1 ∪R5 and ξ = 1R2 ∪R4 , except that the r decay for the bulk
current is not optimised to the r−1−δ and r−3−δ weights for the first and zeroth order terms, respectively,
and the r decay for the boundary current is not optimised to the r−2 weight for the zeroth order term.
(Note that the Boyer–Lindquist r and the r of (2.1) are comparable for large r values so one can compare
directly the r-decay as if the coordinates were the same.) To remedy this, it suffices to add J χ̂V,χ̂w,χ̂q,χ̂$ ,
K χ̂V,χ̂w,χ̂q , respectively to the currents, where (V, w, q, $) here denotes the current of Appendix B.1, χ̂ is
a cutoff supported far away, and is sufficiently small. The resulting currents now indeed satisfy (3.30) and
the first inequality of (3.31), restricted to the exterior, with weights as stated. We note finally that for small
|a|, we may take the χ in the estimate (3.3) proven in [DR10] or [DRSR16] to be identically one outside of
a small neighbourhood of r = 3M . Thus, our ξ indeed satisfies (3.26).
The currents trivially may be extended to the slightly larger bigger domain of Section 2.7.3. Note finally
that the additional positivity statements of (3.31) are easily shown to hold. Rewriting the current in terms
of a single quadruple (V, w, q, $), one easily sees that the boundedness statements (3.15) hold as well. This
completes the proof of Theorem A.1.
−2(X) π µν = X α ∂α (g µν ) − ∂α X µ g αν − ∂α X ν g αµ , (A.29)
77
? ?
which we can write (using g r r
gr? r? = 1 for Kerr in Boyer-Lindquist) as
? ? ? ?
−2(X) π µν = f ∂r? (g µν ) + f g µν gr? r? ∂r? (g r r
) − f g µν gr? r? ∂r? (g r r
) unless µ = ν = r?
? ? ? ? ? ? ? ?
−2(X) π r r
= f gr r
gr? r? ∂r? (g r r
) − 2g r r
∂r? f, (A.31)
so ? ?
−2g r r ∂r? f if µ = ν = r?
(X) µν µν r? r?
−2 π =g · f gr? r? ∂r? (g )+ ? ?
f ∂r? (g µν ) − f g µν gr? r? ∂r? (g r r ) otherwise.
We note also
1 1 √
tr(X) π = gµν (X) π µν = − gµν f ∂r? (g µν ) + ∂r? f = f √ ∂r? g + ∂r? f. (A.32)
2 g
Therefore
(X) µν
π ˜ µ ψ∇
T̃µν = (X) π µν ∇ ˜ ν ψ − 1 (tr(X) π)g µν ∇ ˜ µ ψ∇ ˜ ν ψ − 1 (tr(X) π)Vψ 2
2 2
2 r? r? −1
2
=β g ∂r? f ∂r? (β ψ)
1 2 X h ? ?
i
− β f ∂r? (g µν ) − g µν gr? r? ∂r? (g r r ) ∂µ (β −1 ψ)∂ν (β −1 ψ)
2
µ,ν6=r ?
1 1 √ −1 ˜ µ ψ∇
˜ νψ
− ∂r? f + f √ ∂r? g + f gr? r? ∂r? (gr? r? ) g µν ∇
2 g
1 1 √
− ∂r? f + f √ ∂r? g Vψ 2 . (A.33)
2 g
√
Noting that the determinant g = r2 ∆ 2 2
+a2 Σ sin θ = gr r
? ? r + a sin θ for Kerr in Boyer–Lindquist (t, r? , θ, φ),
? ? 2
(X) µν
π T̃µν = β 2 g r ∂r? f ∂r? (β −1 ψ)
r
1 X h ? ?
i
− β2f ∂r? (g µν ) − g µν gr? r? ∂r? (g r r ) ∂µ (β −1 ψ)∂ν (β −1 ψ)
2
µ,ν6=r ?
1 2r∆ µν 1 1 √
− ∂r? f + f 2 g ∂µ ψ∂ ν ψ − ∂ r ? f + f √ ∂r ? g Vψ 2 .
2 (r + a2 )2 2 g
We now recall from Proposition A.2.1 the definition K̃ X [ψ] = (X) π µν T̃µν [ψ] + X ν S̃ν [ψ] and compute
? ? 2
K̃ X [ψ] = β 2 g r ∂r? f ∂r? (β −1 ψ)
r
1 X h ? ?
i
− f β2 ∂r? (g µν ) − g µν gr? r? ∂r? (g r r ) ∂µ (β −1 ψ)∂ν (β −1 ψ)
2
µ,ν6=r ?
∂r? (β 2 )
1 2 2r∆
− β ∂r? f + f 2 +f g µν ∂µ (β −1 ψ)∂ν (β −1 ψ)
2 (r + a2 )2 β2
1 2 −1 2 1 1 √
− f ∂r? (β V)(β ψ) − ∂r? f + f √ ∂r? g Vψ 2 , (A.34)
2 2 g
78
Recall that, for a given spacetime (M, g) and suitably regular function ψ : M → R, for a given vector
field V , a function w, a one form q and a two form $, the energy current J V,w,q,$ takes the form
Here, for 0 ≤ k ≤ 4, ∗ : Λk M → Λ4−k M denotes the Hodge star operator which satisfies, for all α, β ∈ Λk M,
α ∧ ∗β = g(α, β)dVolM .
where
Divξ = ∗ d ∗ ξ.
In particular it follows that, for any two form $ ∈ Λ2 M and any function ψ ∈ C ∞ (M),
Div ∗ d ψ 2 $ = − ∗ dd ψ 2 $ = 0.
Thus the choice of $ in the current JµV,w,q,$ [g, ψ] never contributes to the associated K V,w,q [g, ψ] or
H V,w [g, ψ]. Moreover q does not contribute to H V,w [g, ψ].
Throughout this section (M, g0 ) denotes Minkowski space (see Section 2.7.1), and
p
r = (x1 )2 + (x2 )2 + (x3 )2 (B.1)
denotes the standard radial coordinate and not the r of (2.1). (Note that (B.1) was denoted r̃ in Section 2.7.1.
Since (2.1) and (B.1) are comparable for large r, the associated r-weighted coercivity properties will be
equivalent.) Recall the (t, r, ϑ, ϕ) and the (u, v, ϑ, ϕ) coordinate systems. The Minkowski metric takes the
form
g0 = −dt2 + dr2 + r2 (dϑ2 + sin2 ϑdϕ2 ) = −dudv + r2 (dϑ2 + sin2 ϑdϕ2 ).
The following basic properties of Minkowski space are used. The spacetime volume form of Minkowski space
can be written
1
dVolM = r2 sin ϑ dt ∧ dr ∧ dϑ ∧ dϕ = r2 sin ϑ du ∧ dv ∧ dϑ ∧ dϕ,
2
and so the Hodge star operator ∗ in particular satisfies
79
B.1 The J V,w,q,$ current
In the case that (M, g0 ) is Minkowski space, the tuple (V, w, q, $) of Section 3.4.1 can be defined as follows.
Consider
V1 = T = ∂u + ∂v , w1 = 0, q1 = 0, $1 = 0,
δ1
V2 = δ1 T = δ1 (∂u + ∂v ), w2 = 0, $2 = − r−1 r2 sin ϑdϑ ∧ dϕ,
q2 = 0,
2
δ2 δ3 δ2 δ3 ∂xµ w3
V3 = 1− (∂v − ∂u ), w3 = 1− , (q3 )µ = − , $3 = 0,
2 (1 + r)δ r (1 + r)δ 2
for appropriate
0 < δ3 δ2 δ1 1, (B.5)
and define
3
X 3
X 3
X 3
X
V = Vi , w= wi , q= qi , $= $i , (B.6)
i=1 i=1 i=1 i=1
and then, for a given ψ, define currents JµVi ,wi ,qi ,$i [ψ] and JµV,w,q,$ [ψ] by (3.10). These satisfy the bound-
edness properties (3.15). (Note that the current JµV2 ,w2 ,q2 ,α2 [g0 , ψ] can be viewed as arising from contracting
the twisted energy momentum tensor, defined in (A.8), with the Killing vector field δ1 T .)
Proposition B.1.1 (The J V,w,q,$ current in Minkowski space). With (V, w, q, $) defined as above, if δ1 ,
δ2 and δ3 are chosen according to (B.5), the current JµV,w,q,$ [ψ] satisfies the coercivity relations (3.17) and
the corresponding K V,w,q [ψ] satisfies the coercivity relation (3.16). More precisely,
δ1 δ1 δ1
JµV2 ,w2 ,q2 ,$2 [ψ]nµΣτ = 2
(∂v (rψ))2 + 2 |∇(rψ)| / 2
+ 2 ιr≤R (∂u (rψ))2
r 4r r
V1 ,w1 ,q1 ,$1 µ 1 2 1 1 2 1
= δ1 Jµ [ψ]nΣτ + δ1 ψ + ψ∂ v ψ + ψ ι r≤R − ψ∂u ψιr≤R ,
4r2 r 4r2 r
δ1 δ1
JµV2 ,w2 ,q2 ,$2 [ψ]nµC = 2 (∂u (rψ))2 + 2 |∇(rψ)| / 2
v 4r 8r
1 2 1
= δ1 JµV1 ,w1 ,q1 ,$1 [ψ]nµC + δ1 ψ − ψ∂ u ψ ,
v 4r2 r
|JµV3 ,w3 ,q3 ,$3 [ψ]nµΣτ | ≤ δ2 C((∂v ψ)2 + |∇ψ| / 2 + (1 + r)−2 |ψ|2 + ιr≤R (∂u ψ)2 ),
|JµV3 ,w3 ,q3 ,$3 [ψ]nµC | ≤ δ2 C((∂u ψ)2 + |∇ψ|
/ 2 + (1 + r)−2 |ψ|2 ),
v
80
and the bulk terms satisfy
It thus follows that the coercivity relations (B.7)–(B.9), along with the property (B.10), hold if δ1 , δ2 and
δ3 are chosen according to (B.5).
The quadruples (Vfar , wfar , qfar , $far ) and (Ṽfar , w̃far , q̃far , $̃far ) of Section 3.5 are defined as follows. Consider
some
0 < δ6 δ5 δ4 δ3 δ2 δ1 1. (B.11)
Noting that − 21 ∂xµ udxµ = − 12 du = (∂v )[ , define
◦
rp−1
(p)
(p)
◦ ◦ ◦ ∂xµ wfar p δ
4
= rp ∂v , rp−2 ∂xµ u,
(p) (p)
$far = −
4
Define then
1 1 1 1
Ṽfar = V, w̃far = w, q̃far = q, $̃far = $,
δ6 δ6 δ6 δ6
where V , w, q, $ are defined by (B.6), so that
(p)
◦
(p)
1 (p)
◦ (p) 1 (p)
◦ (p) 1 (p)
◦ (p) 1
Vfar = Vfar + V, wfar = wfar + w, qfar = qfar + q, $far = $far + $.
δ6 δ6 δ6 δ6
(p)
(p) (p) (p)
Proposition B.2.1 (The Jfar current in Minkowski space). With (Vfar , wfar , qfar , $far ) defined as above, if
(p) (p)
δ1 , . . . , δ6 are chosen according to (B.11), the associated currents Jfar , Kfar , defined by (3.10), (3.11) satisfy
the weighted bulk coercivity properties (3.39) and the weighted boundary coercivity property (3.40). More
precisely, for δ ≤ p ≤ 2 − δ and r ≥ R,
(p) p p
Jfarµ [ψ]nµΣτ & rp−2 |∂v (rψ)|2 + r 2 (∂v ψ)2 + |∇ψ|
/ 2 + r 2 −2 ψ 2 , (B.12)
(p)
81
Proof. Recall again (B.3) and note that
p + 2 p −2 2 rp−1
◦ p + 1 p−2 p
∗ d(ψ 2 $far − 4δ5 r 2 −1 ψ∂u ψ du
(p)
)= r − δ5 r2 ψ −
8 2 2
p + 2 p −2 2 rp−1
p + 1 p−2 p
+ r − δ5 r2 ψ + − 4δ5 r 2 −1 ψ∂v ψ dv.
8 2 2
)µ = − .
4 2
(p)
(p) (p) (p)
◦ ◦ ◦ ◦
The fluxes corresponding to J Vfar ,wfar ,qfar ,$far [ψ] satisfy, for r ≥ R,
(p)
V ◦ ,w◦ ,q ◦ ,$ ◦
(p) (p) (p)
p + 2 p −2 2 p
Jµ far far far far [ψ]nµΣτ = rp−2 |∂v (rψ)|2 − δ5 r 2 ψ + 4r 2 −1 ψ∂v ψ , (B.16)
2
(2 − p) p −2 2 p
= rp−2 |∂v (rψ)|2 + δ5 r 2 ψ − 4r 2 −2 ψ∂v (rψ) ,
2
p
(p)
◦ ◦
V ,w ,q ,$◦ ◦ (p)
r (p)
δ
(p)
p + 2 p −2 2 p
Jµ far far far far [ψ]nµC = |∇ψ| / 2 + 4 rp−2 ψ 2 − δ5 r 2 ψ − 4r 2 −1 ψ∂u ψ , (B.17)
v 4 2 2
and
(p)
(p) (p)
◦ ◦ ◦
K Vfar ,wfar ,qfar [ψ] & prp−3 |∂v (rψ)|2 + (2 − p)rp−1 |∇ψ|
/ 2 + p(2 − p)rp−1 |∂v ψ|2 + (2 − p)rp−3 ψ 2 . (B.22)
(p) (p)
It follows from Proposition B.1.1 that the currents Jfar , Kfar satisfy the weighted bulk coercivity properties
(p)
(B.12)–(B.14) provided δ1 , . . . , δ6 are chosen according to (B.11), and that Hfar satisfies (B.15).
The inequalities (B.10) and (B.15) in particular mean that the r−1 |∇ψ|
/ term on the left hand side of the
assumed inequality (4.30) is superfluous in the case that (M, g0 ) is Minkowski space. This term is estimated
in the proof of Proposition 4.7.2 nonetheless in order to illustrate how it is estimated in the case of Kerr.
82
Proposition C.0.1. Assumption 4.7.1 holds for the classical null condition of Klainerman [Kla86] on
Minkowski space and more generally the class of equations on Kerr considered in Luk [Luk13].
More precisely, if (M, g0 ) is either Minkowski space or a member of the Kerr family and if
satisfies the assumption (C.22), then there exists knull > 0, and for all k ≥ knull , there exists εnull > 0, such
that we have the following.
Let ψ be a smooth function in R(τ0 , τ1 , v) satisfying (4.29) for 0 < ε ≤ εnull . Then for all δ ≤ p ≤ 2 − δ,
X Z 1 1
|LDk ψ| + |LDk ψ| + |∇D
/ k ψ| + |Dk ψ| Dk F
R(τ0 ,τ 0 ,v)∩{r≥R} r r
|k|≤k
Z r
(0+) (0+)
k 2
+ X 8R (τ0 , τ1 )X
(D F ) . k k R
(τ0 , τ1 ),
R(τ0 ,τ 0 ,v)∩{r≥R} 9
and moreover,
X Z 1 1
rp−1 |L(rDk ψ)| + |LDk ψ| + |LDk ψ| + |∇D / k ψ| + |Dk ψ| Dk F
R(τ0 ,τ 0 ,v)∩{r≥R} r r
|k|≤k
Z r q q q
(0) (p) (p) (p) (0)
The estimate of Proposition C.0.1 is only nontrivial for large r and, thus, the main content appears
already around Minkowski space. Accordingly, the proof will be given in detail for the case that (M, g0 ) is
Minkowski space, for the two model nonlinearities
where g/ = r2 γ is the induced round metric and ∇ / its associated connection. See Section C.1. The more
general nonlinearities and the Kerr case are discussed in Section C.2.
L = ∂v , L = ∂u .
The spheres Σ(τ ) ∩ C v are round, g/ = r2 γ denotes the induced round metric and ∇/ its associated connection.
Define Z ∗ Z τ
(0) (0) (p) (p) (p − 1) 1(p − 1)
F
k
∗
:= sup F(τ
k
), Ek ∗ := sup E(τ
k
), Ek 0 := Ek 0 (τ )dτ,
τ0 ≤τ ≤τ1 τ0 ≤τ ≤τ1 τ0
and similarly for k replacing k, and with R subscripts added, etc. This notation will be used throughout
this section.
The main result of this section is the following.
Proposition C.1.1 (Model nonlinearities on Minkowski space). Fix δ ≤ p ≤ 2 − δ and k ≥ 7 and let ψ be
a smooth function in R(τ0 , τ1 , v) ∩ {r ≥ 8R/9} satisfying (4.29).
Then the nonlinearities
83
both satisfy
X Z 1 1 Z
|∂u Dk ψ| + |∂v Dk ψ| + |∇D / k ψ| + |Dk ψ| Dk F + (Dk F )2
R(τ0 ,τ 0 ,v)∩{r≥R} r r R(τ0 ,τ 0 ,v)∩{r≥R}
|k|≤k
s Z ∗ Z ∗ sZ
(0) (0)
(−1 − δ)
(0)
∗
(0)
Z ∗
(δ − 1)
(δ − 1) (0) (−1 − δ)
0 ∗ ∗ 0 0 ∗
. k F ∗R +k
E ∗8R + E
kR
F
k R
+ E
k R
+ E
k R
+ E
kR
F
k R
+ E
k R
9
and
X Z 1 1
rp−1 |∂v (rDk ψ)| + |∂u Dk ψ| + |∂v Dk ψ| + |∇D / k ψ| + |Dk ψ| Dk F
R(τ0 ,τ 0 ,v)∩{r≥R} r r
|k|≤k
sZ sZ s !
Z ∗ q (p − 1)
∗ (0)
Z ∗
(p − 1) (0)
Z
(0) (−1 − δ) (0) (p)
∗(p − 1)
0 0 0
+ k 2
(D F ) . E
kR
F ∗
k R
E
k R
+ F ∗ + E∗ +
k R k
8R E
kR
F ∗
k R
+ E ∗
k R
+ Ek 0R .
R(τ0 ,τ 0 ,v)∩{r≥R} 9
The following properties of Minkowski space will be used in the proof of Proposition C.1.1. First, for any
k and any function ψ, for r ≥ R,
X X
|Dk Lψ| . |LDk ψ|, |Dk Lψ| . |LDk ψ|. (C.1)
e e
|k|≤|k|
e |k|≤|k|
e
dω = sin ϑ dϑ dϕ.
Recall (see Section 2.7.1) that r(u, v) = 21 (v − u). Define v(R) = v(R, u) = 2R + u, so that r(u, v(R, u)) = R,
and similarly define u(R, v) = v − 2R, so that r(u(R, v), v) = R.
Recall the region R(τ0 , τ1 , v). In order to avoid the blow up of constants in Sobolev inequalities, the
region near the corner {u = τ0 } ∩ {r = R} will typically be considered separately from its complement.
Accordingly, define
and
RCorner (τ0 , τ1 , v) := R(τ0 , τ1 , v) ∩ {r ≥ R} ∩ {v 0 ≤ v(R, τ0 + R)}. (C.4)
Note that
3R
r≤ in RCorner (τ0 , τ1 , v). (C.5)
2
Define also
T1 = T1 (v) = min{τ1 , u(R, v)}, and T10 = T1 (v 0 ). (C.6)
84
Proposition C.1.2 (Sobolev inequality on incoming cones C v ). For any τ0 ≤ u ≤ τ1 , v(R, τ0 +R) ≤ v 0 ≤ v,
(ϑ, ϕ) ∈ S 2 (i.e. for any (u, v, ϑ, ϕ) ∈ ŘR (τ0 , τ1 , v)), and any function f : R(τ0 , τ1 , v) → R,
X
Z T10 Z 3
1
X
r|f (u, v 0 , ϑ, ϕ)| . 1 + (τ1 − τ0 )− 2 |Lk f |2 + |Ωi Lk f |2
k≤1 τ0 S2 i=1
3
X 12
|Ωi Ωj Lk f |2 (u0 , v 0 , ϑ, ϕ)r2 dωdu0 .
+
i,j=1
τ1 +τ2
Proof. For simplicity, consider first the case that v 0 ≥ v(R, τ1 ), so that T10 = τ1 . Suppose first that u ≥ 2 .
Let φ be a smooth cut off function such that φ(τ1 ) = 0, φ(τ ) = 1 for τ ≥ τ1 +τ 2 , and
2
4 τ1 + τ2
|φ0 (τ )| ≤ , for τ1 ≤ τ ≤ .
τ2 − τ1 2
Now
∂u φ(u)|f (u, v 0 , ϑ, ϕ)|2 = φ0 (u)|f (u, v 0 , ϑ, ϕ)|2 + 2φ(u)f (u, v 0 , ϑ, ϕ)∂u f (u, v 0 , ϑ, ϕ).
It is in the forms (C.7) and (C.8) that Proposition C.1.2 will be used below.
Proposition C.1.3 (Sobolev inequality on Σ(τ )). For any τ0 ≤ τ ≤ τ1 , v(R, u) ≤ v 0 ≤ v, (ϑ, ϕ) ∈ S 2 , and
any function f : R(τ0 , τ1 , v) → R,
1 Z
X 3
X 3
X 12
|f (τ, v 0 , ϑ, ϕ)| . r−2 |Lk f |2 + |Ωi Lk f |2 + |Ωi Ωj Lk f |2 r2 dωdv 0
Proof. The proof follows from the fundamental theorem of calculus and the standard Sobolev inequality on
S 2 , as in the proof of Proposition C.1.2.
Proposition C.1.3 in particular implies that
r
X (0)
where the region RCorner is defined in (C.4). It is in the form (C.9) that Proposition C.1.3 will typically be
used.
85
C.1.2 The proof of Proposition C.1.1 for the nonlinearity F = ∂u ψ∂v ψ
Proposition C.1.1 for the nonlinearity
F = ∂u ψ∂v ψ,
follows from the following proposition, whose proof is the subject of the present section.
Proposition C.1.4 (Nonlinear error estimates for F = ∂u ψ∂v ψ on Minkowski space). Under the assump-
tions of Proposition C.1.1, for each |k| ≤ k,
Z
X 1 1
/ k ψ| + |Dk ψ| · |Dk1 ∂u ψ| · |Dk2 ∂v ψ|
|∂u Dk ψ| + |∂v Dk ψ| + |∇D
R(τ0 ,τ1 ,v)∩{r≥R} r r
|k1 |+|k2 |≤|k|
Z !
+ |Dk1 ∂u ψ|2 · |Dk2 ∂v ψ|2
R(τ0 ,τ1 ,v)∩{r≥R}
s s
(0)
Z ∗(−1 − δ) (0) (0)
Z ∗(δ − 1) (0)
Z ∗(δ − 1)
(0)
Z ∗(−1 − δ)
. F∗
k R
+ E0
kR
F ∗
k R
+ Ek ∗R + Ek 0R + E ∗8R +
k
E0
kR
F ∗
k R
+ Ek R ,
9
and
X Z
rp−1 |∂v (rDk ψ)| · |Dk1 ∂u ψ| · |Dk2 ∂v ψ|
|k1 |+|k2 |≤|k| R(τ0 ,τ1 ,v)∩{r≥R}
sZ sZ s !
∗(p − 1) q (0)
∗(p − 1) (0) (0)
Z ∗(−1 − δ) (0) (p)
Z ∗(p − 1)
. E0
kR
F ∗
k R
Ek 0R + F∗
k R
E∗
+k 8R + E0
kR
F ∗
k R
+ Ek ∗R + Ek 0R .
9
Proof. The separate terms are estimated individually. We will use (4.29) to replace quartic bounds with cubic
ones without further comment. In view of the fact (C.1), the proof follows from the estimates (C.10)–(C.16)
below.
The notation of Section C.1.1 will be used throughout. Recall, in particular, the regions ŘR and RCorner
defined in (C.3) and (C.4) respectively, the boundedness property (C.5) of r in RCorner , and the T1 and T10
notation defined in (C.6).
The constant in the Sobolev inequality (C.7) blows up as τ1 − τ0 → 0. Accordingly, the cases τ1 − τ0 ≥ 1
and τ1 − τ0 < 1 are typically considered separately. Similarly, the region RCorner , near the corner {u =
τ0 } ∩ {r = R}, is treated separately.
Estimate in the corner region {v 0 ≤ v(R, τ0 + R)}: For any |k̃| ≤ k + 1,
X Z r Z ∗ (0) (−1 − δ)
k̃ k1 k2
|D ψ| · |∂u D ψ| · |∂v D ψ| . k ∗
E 8R Ek 0R . (C.10)
RCorner (τ0 ,τ1 ,v) 9
|k1 |+|k2 |≤|k̃|−1
Indeed, consider some |k1 | + |k2 | ≤ |k̃| − 1. Since k ≥ 7, it follows that either |k1 | ≤ k − 4, or |k2 | ≤ k − 4.
In the former case,
Z
|Dk̃ ψ| · |∂u Dk1 ψ| · |∂v Dk2 ψ|
RCorner (τ0 ,τ1 ,v)
12 Z 12 ∗(−1 − δ)
Z r Z (0)
. sup k1
|∂u D ψ| k̃
|D ψ| |∂v D ψ| k2 ∗
E 8R
. k Ek 0R ,
RCorner(τ0 ,τ1 ,v) RCorner (τ0 ,τ1 ,v) RCorner (τ0 ,τ1 ,v) 9
(C.11)
86
Indeed, suppose |k1 | + |k2 | ≤ |k|. Since k ≥ 7, it follows that either |k1 | ≤ k − 3 or |k2 | ≤ k − 3. If
|k1 | ≤ k − 3, then, for τ1 − τ0 ≥ 1,
Z
|∂u Dk ψ| · |∂u Dk1 ψ| · |∂v Dk2 ψ|
ŘR (τ0 ,τ1 ,v)
Z v Z T10 Z q (0)
Z ∗(δ − 1)
0
sup r|∂u Dk1 ψ| r−1−δ |∂u Dk ψ|2 + r−1+δ |∂v Dk2 ψ|2 r2 dωdudv 0 . k
. F∗ Ek ,
v(R,τ0 +R) u,θ τ0 S2
and, if τ1 − τ0 < 1,
sZ
Z q (0)
∗(−1 − δ)q (0)
|∂u Dk ψ| · |∂u Dk1 ψ| · |∂v Dk2 ψ| . Fk
∗ Ek E ∗.
k
ŘR (τ0 ,τ1 ,v)
87
Indeed, if |k1 | ≤ k − 2, then
Z
|∂v Dk ψ| · |∂u Dk1 ψ| · |∂v Dk2 ψ|
ŘR (τ0 ,τ1 ,v)
Z 12 Z 12
−1
. sup |r∂u D ψ| k1
r |∂v D ψ| k 2
r−1 |∂v Dk2 ψ|2
u,v,θ ŘR (τ0 ,τ1 ,v) ŘR (τ0 ,τ1 ,v)
sZ
q (0)
∗(δ − 1) Z 12
1 0
. 1 + (τ1 − τ0 )− 2 F∗
k
Ek r−1 |∂v Dk ψ|2 ,
ŘR (τ0 ,τ1 ,v)
and, if τ1 − τ0 < 1,
sZ
Z q q
(0) (0)
∗(δ − 1)
k k1 k2 ∗ E∗ 0
|∂v D ψ| · |∂u D ψ| · |∂v D ψ| . F
k k
Ek .
ŘR (τ0 ,τ1 ,v)
. 1 + (τ1 − τ0 )− 2 F
k
∗×
! 12 ! 12
X Z v Z T10 Z Z T10 Z
−1−δ −1+δ
× r k̃
|∂v D ψ|r dωdu 2
r k
|∂v D ψ|r dωdu 2
dv 0
v(R,τ0 +R) τ0 S2 τ0 S2
|k̃|≤|k2 |+3
sZ ! 12
q (0)
∗(−1 − δ) Z v Z T10 Z
− 21 0
. 1 + (τ1 − τ0 ) F
k
∗ E
k
r−1+δ |∂v Dk ψ|r2 dωdudv 0 ,
v(R,τ0 +R) τ0 S2
and, if τ1 − τ0 < 1,
sZ
Z q (0)
∗(−1 − δ)q (0)
k k1 k2 0
|∂v D ψ| · |∂u D ψ| · |∂v D ψ| . F∗
k k
E ∗ Ek .
ŘR (τ0 ,τ1 ,v)
and
s
X Z (0)
Z ∗(−1 − δ) (0) (0)
Z ∗(δ − 1)
0 0
r−1 |Dk ψ| · |∂u Dk1 ψ| · |∂u Dk2 ψ| . F∗ +
k
E
k
∗
F
k
+ Ek ∗ + Ek . (C.14)
|k1 |+|k2 |≤|k| ŘR (τ0 ,τ1 ,v)
88
Indeed, (C.13) and (C.14) follow as in the proof of (C.12), using now the fact that
Z Z ∗ (−1 − δ)
and
Z Z ∗(δ − 1) Z v Z (0)
−1+δ −1 −1+δ −1 0
2 2
|r−1 ∇ψ|
/ 2 + |r−1 ψ|2 r2 dωdv 0 ≤ Ek ∗ .
r |r /
∇ψ| +r |r ψ| ≤ Ek ,
ŘR (τ0 ,τ1 ,v) v(R,τ ) S2
F∗
|∂u Dk1 ψ|2 · |∂v Dk2 ψ|2 . k Ek 0 ,
ŘR (τ0 ,τ1 ,v)
F ∗ Ek ∗ .
|∂u Dk1 ψ|2 · |∂v Dk2 ψ|2 . k
ŘR (τ0 ,τ1 ,v)
Similarly, if |k2 | ≤ k − 3,
Z Z v Z τ1 Z
|∂u Dk1 ψ|2 · |∂v Dk2 ψ|2 . sup |∂v Dk2 ψ|2 |∂v Dk2 ψ|2 r2 dωdudv 0
ŘR (τ0 ,τ1 ,v) v(R,τ0 +R) u,θ τ0 S2
(0)
1
X Z
∗
1 + (τ1 − τ0 )− 2 r−2 |∂v Dk ψ|2 ,
.F
e
k
ŘR (τ0 ,τ1 ,v)
|k|≤|k
e 2 |+3
Indeed, consider first the case τ1 − τ0 ≥ 1. Suppose first that |k1 | ≤ k − 3. Then, if τ1 − τ0 ≥ 1,
Z
rp−1 |∂v (rDk ψ)| · |∂u Dk1 ψ| · |∂v Dk2 ψ|
ŘR (τ0 ,τ1 ,v)
Z 12 Z 12 q Z (0)
∗(p − 1)
0
. sup (r|∂u Dk1 ψ|) rp−3 |∂v (rDk ψ)|2 rp−1 |∂v Dk2 ψ|2 . F∗
k
Ek ,
ŘR (τ0 ,τ1 ,v) RR (τ0 ,τ1 ,v) RR (τ0 ,τ1 ,v)
89
by the Sobolev inequality on the incoming cones (C.7). Similarly, if τ1 − τ0 < 1,
Z
rp−1 |∂v (rDk ψ)| · |∂u Dk1 ψ| · |∂v Dk2 ψ|
ŘR (τ0 ,τ1 ,v)
1
q (0)
Z 21 Z 12 q (0)
q q (p) (0)
. (τ1 − τ0 ) 2 F∗
k
rp−2 |∂v (rDk ψ)|2 rp−2 |∂v Dk2 ψ|2 . F∗
k
Ek ∗ Ek ∗ .
RR (τ0 ,τ1 ,v) RR (τ0 ,τ1 ,v)
using again the Sobolev inequality on the incoming cones (C.8). Similarly, if τ1 − τ0 < 1,
Z
rp−1 |∂v (rDk ψ)| · |∂u Dk1 ψ| · |∂v Dk2 ψ|
ŘR (τ0 ,τ1 ,v)
q Z ! 21 Z ! 12
(0)
− 21
X
∗ p−2 k 2 p−2 k 2
. (τ1 − τ0 ) F r |∂v (rD ψ)| r |∂v D ψ|
e
k
RR (τ0 ,τ1 ,v) RR (τ0 ,τ1 ,v)
|k|≤|k
e 2 |+3
q (0)
q (p)
q (0)
. F ∗ Ek ∗ E ∗.
k k
Z
X 1 1
|∂u Dk ψ| + |∂v Dk ψ| + |∇D/ k ψ| + |Dk ψ| · |∇D
/ k1 ψ| · |∇D
/ k2 ψ|
R(τ0 ,τ1 ,v)∩{r≥R} r r
|k1 |+|k2 |≤|k|
Z ! r Z ∗
(0) (0) (0) (δ − 1)
+ / k1 2 /
|∇D ψ| · |∇D ψ| k2 2
. k ∗ ∗ ∗
E 8R Ek R +
F R +k Ek 0R
R(τ0 ,τ1 ,v)∩{r≥R} 9
and
∗(p − 1)
Z r Z
X (0) (0) (p)
r p−1 k /
|∂v (rD ψ)| · |∇D /
ψ| · |∇D k1
ψ| . k2
F∗
k R
+k 8RE∗ Ek ∗R + Ek 0R .
R(τ0 ,τ1 ,v)∩{r≥R} 9
|k1 |+|k2 |≤|k|
90
Proof. The separate terms are estimated individually, and the proof follows from the estimates (C.17)–(C.21)
below.
The notation of Section C.1.1 will again be used throughout. As in the proof of Proposition C.1.4, the
cases τ1 − τ0 ≥ 1 and τ1 − τ0 < 1 are typically considered separately. Similarly, the region near the corner
{u = τ0 } ∩ {r = R} is treated separately.
Estimate in the corner region {v 0 ≤ v(R, τ0 + R)}: For any |k̃| ≤ k + 1,
X Z r Z ∗ (0) (−1 − δ)
k̃ /
|D ψ| · |∇D k1 /
ψ| · |∇D k2 ∗
E 8R
ψ| . k Ek 0R . (C.17)
RCorner (τ0 ,τ1 ,v) 9
|k1 |+|k2 |≤|k̃|−1
Indeed, (C.17) follows exactly as in the proof of Proposition C.1.4, using the Sobolev inequality (C.9).
The remainder of the proof concerns the region ŘR (τ0 , τ1 , v).
Estimate for ∂u ψ term: First note that, for any |k| ≤ k,
X Z q Z ∗ (0) (δ − 1)
k /
|∂u D ψ| · |∇D k 1 /
ψ| · |∇D k2
F∗
ψ| . k Ek 0 . (C.18)
|k1 |+|k2 |≤|k| Ř(τ0 ,τ1 ,v)
by the Sobolev inequality on the incoming cones (C.7), since p ≥ δ. For τ1 − τ0 < 1,
Z q q q (0) (0) (0)
|∂u Dk ψ| · |∇D
/ k1 ψ| · |∇D
/ k2 ψ| . k
F ∗ Ek ∗ F k
∗.
Ř(τ0 ,τ1 ,v)
|∂v Dk ψ| · |∇D
/ k1 ψ| · |∇D
/ k2 ψ| . F ∗ Ek ∗ ,
k
Ř(τ0 ,τ1 ,v)
using again the Sobolev inequality on the incoming cones (C.7). Similarly for the r−1 ψ and r−1 ∇ψ
/ terms,
using the fact that
Z Z ∗ Z v Z (δ − 1) (0)
r−1 |∇D
/ k ψ|2 + r−3 |Dk ψ|2 ≤ Ek 0 , |r−1 ∇ψ|
/ 2 + |r−1 ψ|2 r2 dωdv 0 ≤ Ek ∗ .
Ř(τ0 ,τ1 ,v) v(R,u) S2
91
which follows as an easy consequence of the Sobolev inequality (C.7).
Estimate for rp−1 ∂v (rψ) term: Finally, for any |k| ≤ k,
X Z q Z (0) (p)
∗(p − 1)
0
rp−1 |∂v (rDk ψ)| · |∇D
/ k1 ψ| · |∇D F ∗ Ek ∗ +
/ k2 ψ| . k Ek . (C.21)
|k1 |+|k2 |≤|k| Ř(τ0 ,τ1 ,v)
Assume again, without loss of generality, that |k1 | ≤ k − 3. Then, using (C.7), for τ1 − τ0 ≥ 1,
Z
rp−1 |∂v (rDk ψ)| · |∇D
/ k1 ψ| · |∇D
/ k2 ψ|
Ř(τ0 ,τ1 ,v)
Z 21 Z 12 q (0)
Z ∗(p − 1)
/ k1 ψ| 0
. sup r|∇D rp−3 |∂v (rDk ψ)|2 rp−1 |∇D
/ k2 ψ|2 . F∗
k
Ek .
Ř(τ0 ,τ1 ,v) Ř(τ0 ,τ1 ,v) Ř(τ0 ,τ1 ,v)
. (τ1 − τ0 )− 2 F∗
k
rp−2 |∂v (rDk ψ)|2 rp−2 |∇D
/ k2 ψ|2 . F∗
k
Ek ∗ Ek ∗ .
Ř(τ0 ,τ1 ,v) Ř(τ0 ,τ1 ,v)
where the A, B indices refer to coordinates ϑ, ϕ (recall the (u, v, ϑ, ϕ) coordinate system of Section 2.7.1), and
Dk are arbitrary constants, the proof of Proposition C.0.1 follows exactly as above, where the bound (4.29)
is now also used to obtain pointwise bounds on ψ through an easy weighted Sobolev estimate which allow
as to invoke (C.22).
Remark C.2.1. Note that, besides the nonlinearities of the classical null condition [Kla86], our class (C.22)
includes for instance also nonlinearities which for large r take the form F = (sin r)∂u ψ∂v ψ. It does not,
however, include even more general examples like F = (sin x)∂u ψ∂v ψ considered recently in [AZ22], due to
the presence of the weighted vector fields Ωi in our set of commutation vector fields Dk .
More generally, if g0 is a metric with asymptotics suitably close to those of Minkowski space, with extra
terms suitably small, then the above proof again applies.
We will consider explicitly the case that g0 is the Kerr metric. Define double null (u, v, θ1 , θ2 ) coordinates
on Kerr, when 0 < |a| < M , in terms of the Boyer–Lindquist coordinates (t, r, ϑ, φ) of Appendix A.1, by
u = t − r∗ , v = t + r∗ ,
92
and θ2 defined by
2M ar
θ2 = φ + h(r∗ , θ1 ), where h satisfies ∂r∗ h(r∗ , θ1 ) = , lim h(r∗ , θ1 ) = 0.
ΣR2 r∗ →∞
where
∆ 4M ar
Ω2 = , b1 = 0, b2 = ,
R2 ΣR2
and
a2 (∂θ1 F )2 (sin2 θ1 − sin2 ϑ)((r2 + a2 )2 − a2 sin2 θ1 ∆)
g/11 = + (∂θ1 h)2 R2 sin2 ϑ,
R2
g/22 = R2 sin2 ϑ, g/12 = g/21 = −R2 sin2 ϑ∂θ1 h.
In particular, one has
L = ∂v + bA ∂θA , L = ∂u .
Moreover T = ∂u +∂v and Ωi = Ωi (θ , θ ) ∂θA for i = 1, 2, 3, for functions Ωi (θ1 , θ2 )A , so that, in particular,
1 2 A
3
X 1 X X
|Dk Lψ| . |LDk ψ| + |Ωi Dk ψ|, (C.26)
e e
r3 i=1
|k|≤|k|
e |k|≤|k|−1
e
hold for large r. Moreover (C.2) remains true. The proof of Proposition C.0.1 in this case then follows just
as in the case where g0 was the Minkowski metric, using now (C.25) and (C.26) in place of (C.1).
We note finally that the condition (C.22) applied to Kerr indeed includes in particular the nonlinearities
considered in [Luk13].
93
References
[AAG16] Y. Angelopoulos, S. Aretakis, and D. Gajic. A vector field approach to almost-sharp decay for
the wave equation on spherically symmetric, stationary spacetimes. arXiv:1612.01565, preprint,
2016.
[AAG20] Y. Angelopoulos, S. Aretakis, and D. Gajic. Nonlinear scalar perturbations of extremal Reissner-
Nordström spacetimes. Ann. PDE, 6(2):Paper No. 12, 124, 2020.
[AAG21] Y. Angelopoulos, S. Aretakis, and D. Gajic. Late-time tails and mode coupling of linear waves
on Kerr spacetimes. arXiv:2102.11884, preprint, 2021.
[AB15] L. Andersson and P. Blue. Hidden symmetries and decay for the wave equation on the Kerr
spacetime. Ann. of Math. (2), 182(3):787–853, 2015.
[ABBM19] L. Andersson, T. Baeckdahl, P. Blue, and S. Ma. Stability for linearized gravity on the Kerr
spacetime. arXiv:1903.03859, preprint, 2019.
[Ali09] S. Alinhac. Energy multipliers for perturbations of the Schwarzschild metric. Comm. Math.
Phys., 288(1):199–224, 2009.
[AMPW17] L. Andersson, S. Ma, C. Paganini, and B. F. Whiting. Mode stability on the real axis. J. Math.
Phys., 58(7):072501, 2017.
[Ape22] M. A. Apetroaie. Instability of gravitational and electromagnetic perturbations of extremal
Reissner-Nordström spacetime. arXiv:2211.09182, preprint, 2022.
[Are11] S. Aretakis. Stability and instability of extreme Reissner-Nordström black hole spacetimes for
linear scalar perturbations I. Comm. Math. Phys., 307(1):17–63, 2011.
[Are12] S. Aretakis. Decay of axisymmetric solutions of the wave equation on extreme Kerr backgrounds.
J. Funct. Anal., 263(9):2770–2831, 2012.
[AZ22] J. Anderson and S. Zbarsky. Global stability for nonlinear wave equations satisfying a generalized
null condition. arXiv:2212.01184, preprint, 2022.
[Ben22] G. Benomio. A new gauge for gravitational perturbations of Kerr spacetimes II: The linear
stability of Schwarzschild revisited. arXiv:2211.00616, preprint, 2022.
[BP73] J. M. Bardeen and W. H. Press. Radiation fields in the Schwarzschild background. J. Math.
Phys., 14(1):7–19, 1973.
[Cha75] S. Chandrasekhar. On the equations governing the perturbations of the Schwarzschild black
hole. Proc. Roy. Soc. A, 343(1634):289–298, 1975.
[Chr86] D. Christodoulou. Global solutions of nonlinear hyperbolic equations for small initial data.
Communications on Pure and Applied Mathematics, 39(2):267–282, 1986.
94
[CK93] D. Christodoulou and S. Klainerman. The global nonlinear stability of the Minkowski space,
volume 41 of Princeton Mathematical Series. Princeton University Press, Princeton, NJ, 1993.
[DHR19] M. Dafermos, G. Holzegel, and I. Rodnianski. Boundedness and decay for the Teukolsky equation
on Kerr spacetimes I: The case |a| M . Ann. PDE, 5(1):Paper No. 2, 118, 2019.
[DHR] M. Dafermos, G. Holzegel, and I. Rodnianski. The linear stability of the Schwarzschild solution
to gravitational perturbations. Acta Math., 222(1):1–214, 2019.
[DHRT21] M. Dafermos, G. Holzegel, I. Rodnianski, and M. Taylor. The non-linear stability of the Schwarz-
schild family of black holes. arXiv:2104.08222, preprint, 2021.
[DL17] M. Dafermos and J. Luk. The interior of dynamical vacuum black holes I: The C 0 -stability of
the Kerr Cauchy horizon. arXiv:1710.01722, preprint, 2017.
[DR05] M. Dafermos and I. Rodnianski. A proof of Price’s law for the collapse of self-gravitating scalar
field. Invent. Math., 162:381–457, 2005.
[DR07a] M. Dafermos and I. Rodnianski. A note on energy currents and decay for the wave equation on
a Schwarzschild background. arXiv:0710.0171, preprint, 2007.
[DR07b] M. Dafermos and I. Rodnianski. The wave equation on Schwarzschild-de Sitter spacetimes.
arXiv:0709.2766, preprint, 2007.
[DR09] M. Dafermos and I. Rodnianski. A new physical-space approach to decay for the wave equation
with applications to black hole spacetimes. In P. Exner, editor, XVIth International Congress
on Mathematical Physics, pages 421–433. World Scientific, London, 2009.
[DR10] M. Dafermos and I. Rodnianski. Decay for solutions of the wave equation on Kerr exterior
spacetimes I-II: The cases |a| m or axisymmetry. arXiv:1010.5132, preprint, 2010.
[DR11] M. Dafermos and I. Rodnianski. A proof of the uniform boundedness of solutions to the wave
equation on slowly rotating Kerr backgrounds. Invent. Math., 185 (3):467–559, 2011.
[DR13] M. Dafermos and I. Rodnianski. Lectures on black holes and linear waves. In Evolution equations,
Clay Mathematics Proceedings, Vol. 17, pages 97–205. Amer. Math. Soc., Providence, RI, 2013.
[DRSR16] M. Dafermos, I. Rodnianski, and Y. Shlapentokh-Rothman. Decay for solutions of the wave
equation on Kerr exterior spacetimes III: The full subextremal case |a| < M . Ann. of Math.
(2), 183(3):787–913, 2016.
[Fan21] A. J. Fang. Nonlinear stability of the slowly-rotating Kerr-de Sitter family. arXiv:2112.07183,
preprint, 2021.
[FM22] M. Facci and J. Metcalfe. Global existence for quasilinear wave equations satisfying the null
condition. arXiv:2208.12659, preprint, 2022.
[Gio20] E. Giorgi. The linear stability of Reissner-Nordström spacetime: the full subextremal range
|Q| < M . Comm. Math. Phys., 380(3):1313–1360, 2020.
[Gio21] E. Giorgi. The Carter tensor and the physical-space analysis in perturbations of Kerr-Newman
spacetime. arXiv:2105.14379, preprint, 2021.
[GKS20] E. Giorgi, S. Klainerman, and J. Szeftel. A general formalism for the stability of Kerr.
arXiv:2002.02740, preprint, 2020.
[GKS22] E. Giorgi, S. Klainerman, and J. Szeftel. Wave equations estimates and the nonlinear stability
of slowly rotating Kerr black holes. arXiv:2205.14808, preprint, 2022.
[HHV21] D. Häfner, P. Hintz, and A. Vasy. Linear stability of slowly rotating Kerr black holes. Invent.
Math., 223(3):1227–1406, 2021.
95
[Hin22] P. Hintz. A sharp version of Price’s law for wave decay on asymptotically flat spacetimes.
Communications in Mathematical Physics, 389(1):491–542, 2022.
[HK20] G. Holzegel and C. Kauffman. A note on the wave equation on black hole spacetimes with small
non-decaying first order terms. arXiv:2005.13644, preprint, 2020.
[HLSW20] G. Holzegel, J. Luk, J. Smulevici, and C. Warnick. Asymptotic properties of linear field equations
in anti–de Sitter space. Comm. Math. Phys., 374(2):1125–1178, 2020.
[Hol10] G. Holzegel. Stability and decay rates for the five-dimensional Schwarzschild metric under biaxial
perturbations. Adv. Theor. Math. Phys., 14(5):1245–1372, 2010.
[HS13] G. Holzegel and J. Smulevici. Decay properties of Klein–Gordon fields on Kerr–AdS spacetimes.
Comm. Pure Appl. Math., 66(11):1751–1802, 2013.
[HS14] G. Holzegel and J. Smulevici. Quasimodes and a lower bound on the uniform energy decay rate
for Kerr-AdS spacetimes. Anal. PDE, 7(5):1057–1090, 2014.
[Hun18] P.-K. Hung. The linear stability of the Schwarzschild spacetime in the harmonic gauge: odd
part. arXiv:1803.03881, preprint, 2018.
[HV16] P. Hintz and A. Vasy. Global analysis of quasilinear wave equations on asymptotically Kerr–de
Sitter spaces. Int. Math. Res. Not. IMRN, (17):5355–5426, 2016.
[HW14] G. H. Holzegel and C. M. Warnick. Boundedness and growth for the massive wave equation
on asymptotically anti-de sitter black holes. Journal of Functional Analysis, 266(4):2436–2485,
2014.
[Joh19] T. W. Johnson. The linear stability of the Schwarzschild solution to gravitational perturbations
in the generalised wave gauge. Ann. PDE, 5(2):Paper No. 13, 92, 2019.
[Keh22] L. Kehrberger. The case against smooth null infinity I: Heuristics and counter-examples. Annales
Henri Poincaré, 23(3):829–921, 2022.
[Kei18] J. Keir. The weak null condition and global existence using the p-weighted energy method.
arXiv:1808.09982, preprint, 2018.
[Kei20] J. Keir. Evanescent ergosurface instability. Anal. PDE, 13(6):1833–1896, 2020.
[Kla86] S. Klainerman. The null condition and global existence to nonlinear wave equations. In Nonlinear
systems of partial differential equations in applied mathematics, Part 1 (Santa Fe, N.M., 1984),
volume 23 of Lectures in Appl. Math., pages 293–326. Amer. Math. Soc., Providence, RI, 1986.
[KS20] S. Klainerman and J. Szeftel. Global nonlinear stability of Schwarzschild under polarized per-
turbations. Annals of mathematics studies, 210(210), 2020.
[Laf22] D. Lafontaine. About the wave equation outside two strictly convex obstacles. Communications
in Partial Differential Equations, 47(5):875–911, 2022.
[LR03] H. Lindblad and I. Rodnianski. The weak null condition for Einstein’s equations. C. R. Math.,
336(11):901–906, 2003.
[LR10] H. Lindblad and I. Rodnianski. The global stability of Minkowski space-time in harmonic gauge.
Ann. of Math., 171(3):1401–1477, 2010.
[LT16] H. Lindblad and M. Tohaneanu. Global existence for quasilinear wave equations close to Schwarz-
schild. arXiv:1610.00674, preprint, 2016.
[LT20] H. Lindblad and M. Tohaneanu. A local energy estimate for wave equations on metrics asymp-
totically close to Kerr. Ann. Henri Poincaré, 21(11):3659–3726, 2020.
96
[LT22] H. Lindblad and M. Tohaneanu. The weak null condition on Kerr backgrounds.
arXiv:2210.10149, preprint, 2022.
[Luk13] J. Luk. The null condition and global existence for nonlinear wave equations on slowly rotating
Kerr spacetimes. J. Eur. Math. Soc. (JEMS), 15(5):1629–1700, 2013.
[Ma20] S. Ma. Uniform energy bound and Morawetz estimate for extreme components of spin fields in
the exterior of a slowly rotating Kerr black hole II: Linearized gravity. Comm. Math. Phys.,
377(3):2489–2551, 2020.
[Mas22a] H. Masaood. A scattering theory for linearised gravity on the exterior of the Schwarzschild black
hole I: The Teukolsky equations. Comm. Math. Phys., 393(1):477–581, 2022.
[Mas22b] H. Masaood. A scattering theory for linearised gravity on the exterior of the Schwarzschild black
hole II: The full system. arXiv:2211.07462, preprint, 2022.
[Mav21] G. Mavrogiannis. Quasilinear wave equations on Schwarzschild-de Sitter. arXiv:2111.09495,
preprint, 2021.
[MMTT10] J. Marzuola, J. Metcalfe, D. Tataru, and M. Tohaneanu. Strichartz estimates on Schwarzschild
black hole backgrounds. Comm. Math. Phys., 293(1):37–83, 2010.
[Mor68] C. S. Morawetz. Time decay for the nonlinear Klein–Gordon equation. Proc. Roy. Soc. A,
306(1486):291–296, 1968.
[Mos16] G. Moschidis. The rp -weighted energy method of Dafermos and Rodnianski in general asymp-
totically flat spacetimes and applications. Annals of PDE, 2(1):6, May 2016.
[Mos18] G. Moschidis. A proof of Friedman’s ergosphere instability for scalar waves. Comm. Math.
Phys., 358(2):437–520, 2018.
[Mos22] G. Moschidis. A proof of the instability of AdS for the Einstein-massless Vlasov system. Inven-
tiones Mathematicae, pages 1–206, 2022.
[MS05] J. Metcalfe and C. D. Sogge. Hyperbolic trapped rays and global existence of quasilinear wave
equations. Inventiones Mathematicae, 159(1):75–117, 2005.
[NP62] E. Newman and R. Penrose. An approach to gravitational radiation by a method of spin
coefficients. J. Math. Phys., 3(3):566–578, 1962.
[Pas19] F. Pasqualotto. Nonlinear stability for the Maxwell-Born-Infeld system on a Schwarzschild
background. Ann. PDE, 5(2):Paper No. 19, 172, 2019.
[PI98] F. Pretorius and W. Israel. Quasispherical light cones of the Kerr geometry. Class. Quant.
Grav., 15:2289–2301, 1998.
[RW57] T. Regge and J. A. Wheeler. Stability of a Schwarzschild Singularity. Phys. Rev., 108:1063–1069,
1957.
[Sbi15] J. Sbierski. Characterisation of the energy of Gaussian beams on Lorentzian manifolds: with
applications to black hole spacetimes. Anal. PDE, 8(6):1379–1420, 2015.
[SR15] Y. Shlapentokh-Rothman. Quantitative mode stability for the wave equation on the Kerr space-
time. Ann. Henri Poincaré, 16(1):289–345, Jan 2015.
[SRTdC20] Y. Shlapentokh-Rothman and R. Teixeira da Costa. Boundedness and decay for the Teukol-
sky equation on Kerr in the full subextremal range |a| < M : frequency space analysis.
arXiv:2007.07211, preprint, 2020.
[Sto17] J. Stogin. Nonlinear Wave Dynamics in Black Hole Spacetimes. PhD thesis, Princeton Univer-
sity, 2017.
97
[TdC20] R. Teixeira da Costa. Mode Stability for the Teukolsky Equation on Extremal and Subextremal
Kerr Spacetimes. Comm. Math. Phys., 378(1):705–781, 2020.
[Teu73] S. A. Teukolsky. Perturbations of a rotating black hole. I. Fundamental equations for gravita-
tional, electromagnetic, and neutrino-field perturbations. Astrophysical J., 185:635–648, 1973.
[Whi89] B. F. Whiting. Mode Stability of the Kerr black hole. J. Math. Phys., 30:1301–1306, 1989.
[WZ11] J. Wunsch and M. Zworski. Resolvent estimates for normally hyperbolic trapped sets. Ann.
Henri Poincaré, 12(7):1349–1385, 2011.
[Yan13] S. Yang. Global solutions of nonlinear wave equations in time dependent inhomogeneous media.
Archive for Rational Mechanics and Analysis, 209(2):683–728, Aug 2013.
98